A systematic review of evidence on malignant spinal metastases: natural history and technologies for identifying patients at high risk of vertebral fracture and spinal cord compression

Implications for research

There is a need for:

1. Prospective randomised designs of the clinical effectiveness and cost-effectiveness of identification and subsequent treatment of patients at high risk of vertebral collapse and SCC. These trials should be undertaken for diagnostic methods such as bone scintigraphy and particularly for serial MRI, to identify patient groups who are most likely to benefit from early detection and treatment, and the value of, and optimal frequency of MRI screening for populations.

2. Service Delivery and Organisation research on MRI and scanning (in tandem with research studies on use of MRI to monitor progression) in order to understand best methods for maximising use of MRI scanners (e.g. to investigate variation in need, and optimal location, throughput and staffing, etc.).

3. Investigation of prognostic algorithms designed to calculate the probability of a specified event using high-quality prospective studies, involving defined populations, randomly selected and clearly identified samples, and with blinding of investigators.

4. Higher-quality prospective studies to investigate and confirm previous findings on risk factors for progression or spinal collapse, as opposed to survival. These could usefully feed into work on prognostic algorithms.

5. Methodological research to improve prognosis research.

Implications for clinical practice

The major factors that should be taken into account when considering a patient for further investigation and potential treatment when at risk of SCC, progression or spinal collapse have not altered from those identified in 2008 NICE guideline 75.

Treatment of patients with spinal metastases and prevention of metastatic spinal cord compression

Patients with painful spinal metastases should be offered conventional analgesics, i.e. non-steroidal anti-inflammatory drugs. Those patients with intractable pain should be considered for specialist pain care that includes invasive procedures and neurosurgical interventions. Patients with spinal metastases from breast and prostate cancer should be offered bisphosphonates to alleviate pain and reduce the risk of pathological fracture/collapse of the spine. Those patients with non-mechanical spinal pain should be given single-fraction palliative radiotherapy. This should also be considered in those who are completely paralysed. In asymptomatic patients, radiotherapy should not be administered.

Two vertebral augmentation techniques, vertebroplasty and kyphoplasty, should be considered in those with mechanical spinal pain resistant to conventional analgesics and no evidence of MSCC or spinal instability.

Surgery should be preferred when there is evidence of progressive disease mainly to prevent MSCC. It should also be considered in those with spinal metastases and mechanical pain resistant to conventional analgesics and in those with evidence of spinal instability.

Treatment of threatened spinal cord in patients with metastatic spinal cord compression

In patients with severe mechanical pain suggestive of spinal instability or those with neurological symptoms or signs suggestive of MSCC, the spine should be stabilised. Patients should be monitored regularly, especially during sitting from supine to 60 degrees. If patients continue to deteriorate, then they should revert back to the lying position or to the position in which there is minimal pain/neurological symptoms. In those patients not suitable for definitive treatment, the aim of the treatment should be helping the patient to achieve a comfortable position and mobilisation. This is usually achieved by using orthoses.

Corticosteroids should be given to all patients with MSCC unless contraindicated. Dexamethasone at 16 mg as a loading dose should be given followed by a short course of 16 mg dexamethasone daily until definitive treatment is employed. After definitive treatment, the dose of dexamethasone should be reduced gradually over 5–7 days and then stopped. In those patients in whom symptoms have deteriorated, the dose of dexamethasone can be increased temporarily.

Definitive treatment of metastatic spinal cord compression

The definitive treatment should be given as early as possible, ideally within 24 hours of the diagnosis of MSCC. Before this, diagnosis of primary location of the tumour should be made. In addition, an attempt should be made to study the extent of the disease. A scoring system such as the Tokuhashi scoring system and the American Society of Anesthesiologists grading for overall patient condition should be used to assess whether surgery is appropriate.

Corticosteroids

Intravenous or oral corticosteroids have been found to provide improvement or resolution of neurological symptoms and pain in patients with epidural spinal metastases. 63 It should be noted that there is no standard dosage regimen for corticosteroids. They are often used before surgery. 63

In patients with MSCC undergoing surgical decompression, corticosteroids are often used in combination with radiotherapy. 74

Bisphosphonates and denosumab

Bisphosphonates are known to impair osteoclastic activity and so they reduce tumour-related resorption of bone. 10,57 Currently, bisphosphonates are used to alleviate metastatic bone pain and to reduce SREs such as pathological fractures, hypercalcaemia and MSCC. Bisphosphonates are also used to reduce the frequency of surgery and radiation therapy. 10,57 Bisphosphonates such as pamidronate (Aredia^®; Novartis Pharmaceutitcals Corporation), clodronate (Bonefos^®, Clasteon^®, Loron^®; Bayer), ibandronate (Bondronat^®; F. Hoffmann-La Roche Ltd), alendronate (Fosamax^®; Merck Sharp & Dohme Corporation) and zoledronate (Aclasta^®; Novartis) have all been found to be effective in the treatment of hypercalcaemia. 57 Although radiotherapy is the main treatment for reducing bone pain, bisphosphonates can be used as an alternative therapy, which in turn will considerably reduce the frequency of radiotherapy. 10,57 The effect of bisphosphonates on pain is not dependent on the nature or type of the tumour (i.e. sclerotic or lytic). 57 The efficacy of these drugs has been seen in breast cancer, multiple myeloma and other osteolytic metastases. 57 Although bisphosphonates have been found to be effective in preventing skeletal-related complications, they are not so effective in reducing pain in patients with prostate cancer. 77

Recently, monoclonal antibody therapy with denosumab, a specific inhibitor of RANKL, has been found to be effective in delaying and preventing SREs. 77

Chemotherapy

The benefits of chemotherapy are limited in spinal metastases, as patients are usually at a late stage of disease. 57 Chemotherapy can be given on its own or in combination with surgery and hormonal therapy. 10

Radioisotopes

Radioisotopes are administered systematically and act as local radiation therapy to the spine. 10 Radioisotopes include strontium-89 and rhenium-186. Although radioisotopes are found to reduce pain in patients with spinal metastases, these can cause irreversible bone marrow suppression and, for this reason, they are recommended for use in those with good marrow function and in whom no other treatment is available. 10

Hormonal therapies

Hormonal therapies are a major treatment modality for metastatic breast and prostate cancer. As an example, a new drug, abiraterone (Zytiga^®; Janssen), has recently been developed which has improved outcomes in men with metastatic castration-resistant prostate cancer. 78,79

Prognosis

Several types of prognostic studies have been undertaken to explore the prognosis of spinal metastases. These studies will be the focus of this current short report. Prognostic studies serve several purposes, for example to inform choice of potential pre-emptive intervention(s) so as to avoid or delay more radical surgical intervention; to bring forward radical interventions before patients' health deteriorates to the extent that they are no longer suitable candidates for interventions; and to categorise patients into those more or less suitable for earlier or later radical intervention.

Prognostic studies comprise four types:

• attempts to determine the risk factors that allow prediction of overall survival (e.g. scoring schemes such as those of Tokuhashi et al. 85 and Tomita et al. 86)

• the identification of patients most suitable for surgical intervention; some of these studies are specific for metastases derived from particular primary tumours (e.g. lung, breast)

• attempts to identify risk factors important in determining the survival of patients after surgical intervention for SCC and/or vertebral compression fracture(s)87 (e.g. vertebrectomy and reconstruction, vertebroplasty, kyphoplasty, radiofrequency ablation)

• assessment of risk factors using clinical or imaging technologies for progression of metastatic spinal metastases to SCC and/or to vertebral compression fracture(s). 88,89

Early studies by Yamashita et al. 90 documented longer survival in patients with spinal or pelvic metastatic cancer lesions compared with those with appendicular lesions or both. Tokuhashi et al. 81 developed a scoring system involving six parameters to determine survival after surgery for metastatic spinal tumours: (1) general condition; (2) number of vertebral metastases; (3) number of metastases to internal organs; (4) number of metastases to extraspinal bone; (5) primary site; and (6) severity of spinal cord injury. Scores of 9 out of a possible 12 indicated a good prognosis for patients whereas scores < 5 indicated a worse prognosis. 81 Tomita et al. 86 developed a similar scoring system based on (1) primary tumour site, (2) presence of visceral metastases and (3) number of bone metastases. In contrast to Tokuhashi et al. ,81 in this system, a lower score indicates a better prognosis. 86

van der Linden et al. 91 analysed response to radiotherapy in a cohort of patients with painful spinal metastases and without neurological impairment. Patient characteristics such as Karnofsky performance score, primary tumour site, number of visceral metastases, etc., were studied for their prognostic value in predicting survival. The points were awarded as follows: (1) 2, 1 and 0 points were given for Karnofsky performance score of 80–100, 50–70 and 10–40, respectively; (2) 3, 2, 1 and 0 points were given for breast cancer, prostate cancer, lung cancer and other types of cancer, respectively; and (3) in the presence of visceral metastases 1 point was given, and 0 points if they were absent. Three prognostic groups were formed: Group A with scores between 0 and 3, Group B with scores between 4 and 5, and Group C with a total score of 6. The median overall survival in Groups A, B and C was found to be 3 months, 9 months and 18.7 months, respectively. Patients in Group C had breast cancer with good performance and no metastases to organs. 91

Sioutos et al. 92 studied a cohort of patients with spinal metastases from solid tumours and epidural compression of the spinal cord who underwent surgical decompression of the spinal cord and radiotherapy. Patient characteristics such as anatomical site of primary carcinoma, preoperative neurological deficit, extent of disease, number of vertebral metastases, site of cord compression and age were explored if they predicted survival. In the study, it was found that patients with renal cell carcinoma survived longer than those with breast, prostate, lung or colon cancer. Patients with single vertebral body metastasis survived comparatively longer than those with multiple vertebral body metastases. The presence of leg strength between 0/5 and 3/5, lung or colon cancer, and multiple vertebral metastases all had a negative impact on survival; however, factors such as extent of disease, age and location of tumour had no apparent impact on overall survival of patients. 92

Ambulatory status, age < 60 years and single vertebral segment involvement have also been found to be independent predictors of good outcome. 92–95 Furthermore, Weigel et al. 95 reported a significant association between a postoperative Karnofsky scale and duration of survival.

Bauer and Wedin96 studied survival of patients with spinal metastases after surgery. The survival of the patient was found to be associated with metastatic load, location of tumour and presence of pathological fracture. On multivariate regression analysis, some factors such as pathological fracture, metastasis to brain or viscera, and lung cancer were found to be negative prognostic factors while single skeletal metastases and breast or kidney cancer were positive variables. 96

All of these scoring systems relate to survival, which is not one of the outcome measures included in this current review. According to a recent review, they have recently been assessed as having limited predictive value. 8

Study design

Randomised controlled trials (RCTs), systematic reviews, prospective or retrospective case series, cohort or case–control studies (case studies were excluded).

Population

Adult patients with vertebral metastases at risk of developing (or who have developed) MSCC, vertebral collapse or progression of vertebral collapse.

Intervention/technologies

Diagnostic/prognostic methods, including clinical features and/or imaging technologies [MRI, CT, PET, technetium-99m (⁹⁹Tc^m) scintigraphy, radiography], and natural history.

Comparator

None or another diagnostic/prognostic method.

Outcomes

Spinal cord compression, vertebral compression, vertebral collapse or progression of vertebral collapse.

Treatment

The reporting of the principal treatment and diagnostic or prognostic tool and also the proportion of patients who had had treatment were evaluated.

Recruitment dates

The time period during which patients were recruited was established.

Baseline characteristics

Known prognostic factors were included, for example whether there were differences between studies in terms of the stages of the cancers.

Relevant aim

The aim of this study was to identify risk factors for occult subarachnoid space (SAS) compression or SCC in patients judged to be at risk according to clinical, radiography or bone scan parameters. Occult SAS compression or SCC was established using MRI.

Design and method

This study investigated outpatients with metastatic prostate cancer at a single Canadian hospital (n = 68). All had previous evidence of spinal metastases, but had no neurological indication or signs of SAS compression or SCC; 64 out of 68 had received continuous hormone therapy and 61 had hormone-refractory disease (indicated by rising PSA levels). The authors described the sample as cross-sectional; however, as patients were approached at the physicians' discretion, the sample was probably one of convenience. All patients were examined by bone scintigraphy within 1 week of study entry. Follow-up ranged from 1 to 47 months (median 8 months). Thirty patients underwent plain radiographic examination: 22 for back pain and eight at the physicians' discretion. The timing of radiography was not reported.

Results

MRI was used to identify patients with occult SAS compression/SCC; the timing of the MRI was not stated. The criteria used to establish occult compression are shown in Box 1.

Occult compression was identified in 22 out of 68 patients; all cases were due to direct extension of tumour from the vertebral body. Ten patients had frank compression of the cauda equina or the spinal cord, and 12 of the SAS alone. Nine of 22 had compression at two separate vertebral levels. The disposition of compressions was cervical in three patients, thoracic in 20 patients and lumbar in eight patients. Plain radiographs were not informative for detection of occult compression.

Clinically evident SCC developed during follow-up in 4 of the 46 patients in whom no occult compression was apparent at MRI (the authors quote actuarial risk at 1 and 2 years using Kaplan–Meier analysis; however, with only four events it is unlikely that this analysis is meaningful). The 22 patients with occult compression were treated with appropriate radiotherapy; the post-MRI occurrence of neurological compression in these 22 patients was not reported.

Logistic regression was used to identify risk factors for occult SAS compression/SCC. Of the candidate factors examined in univariate regression, haemoglobin, duration of continuous hormone therapy before study entry and bone scan extent of disease (EOD) score (extent of disease according to number of bone metastases according to the Soloway et al. method136) were significantly associated with occult compression (p = 0.04, p = 0.03 and p = 0.015, respectively), whereas no association was found for Gleason score, alkaline phosphatase, PSA, prostatic acid phosphatase, presence of back pain or use of narcotic analgesics. In multivariate regression, only EOD and duration of hormone therapy were significantly associated with occult compression (p = 0.02 and p = 0.04, respectively).

Author conclusion

Heavy load of spinal metastases, as indicated by scintigraphy, and duration of continuous treatment with hormone therapy are predictive factors for presence of occult SAS compression or SCC. Such patients, although lacking neurological abnormality and signs of compression, would probably benefit from early MRI for occult compression, which if positive for compression should be followed by radiotherapy treatment before the development of symptomatic compression.

Reviewer conclusions

Patients with a high-risk bone scan may benefit from MRI of the spine aimed at early detection and treatment of occult SAS compression/SCC. The reported results are as would be intuitively expected, so the more spinal metastases that are present and the longer a patient is at risk, the greater the chance of clinically occult SCC. The time a patient is on hormone therapy is a proxy for how long he or she is at risk of occult compression. The quantitative estimates of risk probably do not add much value to this conclusion other than suggesting that spinal load is more influential than time at risk; it can be hypothesised that time at risk interacting with the individual patient's propensity for metastases to reach the spine will govern the spinal load. What this study does not address is the probability that occult SCC becomes patently symptomatic SCC, and how long after occult SCC is detected this occurs.

Relevant aim

The aim of this study was to identify risk factors for recurrence of SCC at an old or new site in prostate cancer patients with previous diagnosis of SCC.

Design and method

The main focus of this retrospective study was to identify prognostic factors for survival and for good response to therapy, after diagnosis of SCC in prostate cancer patients with SCC treated at the Royal Marsden Hospital between 1984 and 1992. Sixty-nine patients with SCC were identified from a review of medical records of (1) participants in a previous study of hormone-resistant prostate cancer; (2) those who had undergone spinal MRI; and (3) those who had undergone spinal irradiation. The total number of records reviewed was not stated. SCC was established by myelography or MRI in 66 patients and from plain radiographs in three patients. No information was provided about patients with negative assessments for suspected SCC. Thirteen of 69 patients had SCC at presentation; the median time from diagnosis of prostate cancer to detection of SCC in the remaining 56 patients was 586 days. Most patients (n = 52) had received hormone therapy. Evidence of vertebral collapse at the site of cord compression was present in 24 patients. MRI identified more patients with multiple sites of SCC than did myelography. Fifty-seven patients were given radiotherapy after SCC diagnosis and 13 received surgery.

Results

Neurological relapse (from various causes including 13 second occurrences of SCC) was observed in 20 out of 69 patients. A second SCC at the same site occurred in eight patients and at a new site in five patients.

None of the following potentially predictive factors were associated with the occurrence of neurological relapse: presenting characteristics; haemoglobin; the number of lesions evident by bone scan; hormonal status or method of diagnosis; radiation dose for first SCC. The paper provides a Kaplan–Meier analysis of the cumulative probability of neurological relapse. The methodology used for this was unclear and no ‘at risk’ table was provided; one interpretation is that all 69 patients were included and many were censored at time of death if no relapse had occurred.

Author conclusions

No significant factor was identified for risk of future relapse. An early improvement in motor power is a strong predictor of subsequent functional improvement. MRI detects additional sites of asymptomatic SCC which makes it the investigation of choice.

Reviewer conclusions

No significant predictive factor was identified for risk of future relapse (i.e. second SCC) but the sample was so small that there was little power in the analysis. The actuarial analysis of time to relapse was difficult to interpret because of a lack of methodological detail and ambiguity about the equivalence of SCC and neurological relapse.

Relevant aim

The aim was to determine and analyse, with reference to primary tumour stage and differentiation at diagnosis, the interval between primary diagnosis and SCC, the interval between radiographic evidence of bony metastasis and cord impingement, and the survival period after spinal cord compromise.

Design and method

Forty-one patients with biopsy-proven adenocarcinoma of the prostate who presented with MSCC, or who subsequently developed MSCC, were identified from a total of 611 prostate cancer patients seen at a single centre over a period from 1975 to 1983. Mean and median age was 68 (range 50–90) years. Primary tumours were classified according to a modified Gleason system and patients were classified according to the Fowler–Whitmore staging system. SCC was established by myelography (n = 36) and by clinical findings (n = 5), and by bone scan- or plain radiography-detected lesions at the level of compression. Of the 41 patients with SCC, 3, 11 and 27 were classified as stages A, B and C at the time of diagnosis of the primary tumour. The spinal locations of the SCCs were reported as follows: cervical, two patients; thoracic, 21 patients; lumbar, 14 patients; cervical and thoracic, one patient; cervicothoracic junction, one patient; thoracic and lumbar, two patients.

Results

The median time between primary diagnosis and SCC was 24 months, with 5 out of 41 patients presenting with SCC; there was no clear relationship with tumour grade at diagnosis. The median time from detection of bone metastases to SCC was 15.5 months, with 6 out of 41 patients having bone metastases first observed at the diagnosis of SCC. A second SCC developed in six patients during follow-up, at 6, 6, 8, 19, 21 or 23 months after the first SCC.

Author conclusions

Overall, tumour stage and differentiation were poor predictors of prognosis once a diagnosis of cord compression was established. MSCC secondary to adenocarcinoma of the prostate most frequently occurs in a thoracic location in patients with poorly differentiated disease at diagnosis. The mechanism of cord involvement appears to begin with osseous vertebral metastasis, progressing to extradural compromise with a median interval that is independent of tumour grade. The prognosis following spinal cord involvement remains dismal in the majority of cases.

Reviewer conclusions

This paper did not look at predictive factors of SCC other than tumour grade at diagnosis of primary tumour and this had no detectably significant influence on median time to SCC detection. Kaplan–Meier analysis was not used.

Relevant aim

The study aimed to determine if high-resolution bone scintigraphy at the time of diagnosis of hormone-refractory metastatic prostate cancer has added prognostic value compared with prevailing PSA concentrations and tumour staging (Gleason grading) for SCC-free survival. The authors were also interested in prognosis for overall survival.

Design and method

This was a retrospective study of 84 patients with histologically confirmed diagnosis of hormone-resistant metastatic prostate cancer (treatments had included orchidectomy, luteinising hormone-releasing hormone agonist or antiandrogens). Hormone treatment was stopped for 23 patients. Patients were followed up till death. Tumours were assessed according to Gleason grading. Various criteria were used to establish hormone refractoriness. Palliative care treatments included radiotherapy, radionuclide therapy (⁸⁹Sr), nitrogen-containing bisphosphonate olpadronate and analgesia.

Whole-body high-resolution bone scintigraphy was undertaken using ⁹⁹Tc^m-labelled methylene-diphosphonate. Bone scans were scored according to the Soloway scoring system; in addition, metastasis of vertebrae was classified according to degree of involvement as total or partial. Scintigrams were interpreted by two independent observers. Observations were related to subsequent development of SCC established clinically according to impaired motor or sensory function confirmed by MRI or CT at the appropriate spinal level. ‘Skeletal event-free survival’ was defined as survival without SCC.

Kaplan–Meier analysis of overall survival and of SCC-free survival was undertaken with Cox regression to investigate the relationship between the relative risk (RR) of SCC according to variables including PSA (log-transformed), serum alkaline phosphatase (log-transformed), Soloway grade, age, degree of vertebral involvement and Gleason score.

Results

Mean survival after hormone refractoriness was 8.6 [standard deviation (SD) 10.6] months. Twenty of 84 patients developed SCC 3 days to 10 months after refractoriness was established. Six patients experienced SCC at lumbar level, and 14 at thoracic level; four of the latter also had SCC at another level, two at lumbar level and two at cervical level.

Mean Gleason score was 7.5. When Gleason score was dichotomised to ≥ 7 or < 7, the former was found to be significantly associated with shorter SCC-free survival and overall survival (median 6.1 vs. 12.3 months; p < 0.05). Median overall survival of patients with Gleason scores ≥ 7 and < 7 was 6.8 months and 12.7 months, respectively (p < 0.03). RR (Gleason ≥ 7 vs. < 7) was 1.89 (95% CI 1.02 to 3.53) for mortality and 1.76 (95% CI 0.95 to 3.28) for SCC-free survival. RRs remained significant after adjusting for confounders: RR = 2.33 (p = 0.013) for mortality and RR = 2.37 (p = 0.003) for SCC.

Serum PSA and alkaline phosphatase activity were elevated in all patients (mean values of 511 ± 1035 μg/l and 402 ± 503 U/l, respectively). Log-transformed PSA concentrations were significantly predictive of SCC-free survival (RR = 1.21, 95% CI 1.07 to 1.36; p = 0.003) and survival (RR = 1.17, 95% CI 1.04 to 1.32; p = 0.01) but log-transformed serum alkaline phosphatase activity and age were not.

The unadjusted RR for SCC was significantly associated with Soloway grade (p = 0.031), i.e. patients with heavier metastatic skeletal load were more likely to sustain SCC (the reported results are summarised in Table 11). However, after adjustment for confounders (PSA and alkaline phosphatase concentrations and age), statistical significance was greatly reduced to p = 0.35 (similarly, for survival, p = 0.008 became p = 0.09 after adjustment).

The unadjusted RR for SCC-free survival among Soloway grade 4 patients was significantly greater than that for grade 1 patients (this also applied for overall survival).

For the ‘new method’ of assessing total or partial vertebral involvement at progression, the sensitivity and specificity were 0.90 and 0.94, respectively (based on 2 × 2 table values shown in Table 12 derived from table 2 of published paper). The positive predictive value for a totally involved vertebra was therefore 82% and the positive likelihood ratio (LR) was 14.4. The SCC pre-test probability of ≈ 0.24 is raised to a post-test probability of ≈ 0.82 by a positive total involvement test.

Author conclusions

The data demonstrate that bone scintigraphy performed at the time of development of refractoriness to hormone therapy is of high predictive value for inherent risk of subsequent SCC.

Reviewer conclusions

There is no indication of how the 84 patients were selected. It is possible that different patients received various treatments likely to influence the probability of SCC (e.g. bisphosphonates), but it is not clear if these were accounted for in Cox regression analyses. Although the ‘total involvement of vertebra’ according to scintigraphy appeared to be highly sensitive and specific for subsequent SCC, the study lacks sufficient rigour to be confident of this result; in particular, participant selection was unclear, progression criteria were not defined precisely and no details were given of the method of discriminating total from partial vertebral involvement except that two independent assessors were involved. The frequency of investigator disagreements was not reported, and the level of concordance and how investigator disagreement was handled were not mentioned. It is not altogether clear whether the assessment was conducted before or after SCC was determined to have occurred, and whether scintigraphy assessors and MRI/CT assessors were reciprocally blind to each other's results.

Relevant aim

The aim of the study was to identify clinical factors that predict a high risk for SCC in metastatic prostate cancer patients with MRI-suspected overt or occult SCC who have no functional neurological deficit.

Design and method

This was a retrospective study based on a single institution's medical records of 570 consecutive patients with prostate cancer who underwent MRI of the spine between January 2001 and May 2005. Patients with skeletal metastases were included if their MRI indicated SCC in the absence of neurological deficit. Patients were excluded if they had experienced previous SCC. In all, 150 patients were included. Their median age was 69 (range 50–88) years, 112 out of 150 were hormone refractory, median time since diagnosis was 41.3 (range 3.13–213) months and from start of hormone treatment was 26.8 (range 0.1–157.5) months. Gleason scores ranged from 6 to 10 (G6, n = 23; G7, n = 36; G8, n = 26; G9, n = 29; G10, n = 9). Back pain was present in 72% of patients.

Whole-spine MRI was conducted and patients were classified as (1) ‘overt SCC’, defined as involvement or compression of either the spinal cord or the cauda equina by an epidural or an intramedullary mass lesion; (2) ‘occult SCC’, defined as metastatic disease causing impingement, indentation or loss of definition of the thecal sac; or (3) no SCC. Categories (1) and (2) were considered together as radiological SCC (rSCC).

Binary univariate and multivariate logistic regression were used to identify independent clinical risk factors for rSCC.

Results

Of the 150 patients, 41 (27.33%) had rSCC, 24 (16%) had overt rSCC and 17 (11.3%) had occult rSCC. Seven patients had rSCC at multiple non-contiguous sites; 20 had compression in the thoracic spinal level and 21 in the lumbosacral region.

On univariate analysis, significant determinants of rSCC were found to be extensive bone metastases (six or more bone lesions; p = 0.005) and back pain (p = 0.002), whereas age (p = 0.97), time from diagnosis (p = 0.52), metastasis at diagnosis (p = 0.535), Gleason score (p = 0.34), hormone refractory status (p = 0.158), time from starting hormonal treatment (p = 0.96) and PSA at the time of MRI (p = 0.855) did not predict rSCC. Univariate regression results are summarised in Table 13.

On multivariate analysis, back pain [odds ratio (OR) 5.1, 95% CI 1.44 to 18.25; p = 0.012] and extensive bone metastasis (OR 2.9, 95% CI 1.01 to 8.35; p = 0.047) were significant independent predictors of rSCC. One variable, PSA at the time of MRI (median PSA 402 vs. 98 ng/ml), was significantly different in the patients who had overt SCC and those who had occult SCC [hazard ratio (HR) 1.005, 95% CI 1.001 to 1.009; p = 0.014].

Author conclusions

A significant proportion (27.3%) of patients with metastatic prostate cancer may harbour overt or occult SCC in the absence of functional neurological deficit. MRI of the spine for the early diagnosis of SCC may be considered useful in patients with extensive skeletal metastasis and back pain.

Reviewer conclusions

Magnetic resonance imaging of the spine in patients with extensive skeletal metastasis and back pain but lacking neurological deficit may lead to early detection of SCC. Since 72% of the selected population had back pain, and rSCC may occur in patients without back pain (although relevant data were not provided), it appears that MRI of all patients with extensive skeletal metastases might represent a factor for consideration.

Relevant aim

The authors' stated aims were to determine the optimal frequency of MRI of the spine required to detect clinically occult rSCC and to determine the incidence of neurological deficit in patients with metastatic prostate cancer. The rSCC was defined as involvement or compression of either the spinal cord or the cauda equina by an epidural or an intramedullary mass/lesion, or metastatic disease causing impingement, indentation or loss of definition of the thecal sac.

Design and method

This was a retrospective study based on a single institution's medical records of 570 consecutive patients with prostate cancer who underwent MRI of the spine between January 2001 and May 2005. Initial MRI was requested by the physician when patients were considered at high risk of SCC. These appear to be the same patients as those in Venkitaraman et al. 2007. 132 From these 570, 130 patients (median age 70 years, range 50–88 years) with castration-resistant disease and skeletal metastases in whom MRI was indicative of suspected clinically occult SCC (i.e. an absence of neurological deficit) were included. Patients were excluded if they had experienced previous SCC. Median times since diagnosis and the start of hormone treatment were 1355 (range 219–6412) days and 917 (range 100–3332) days, respectively. The selection criteria differed from those in the previous 2007 study in that some patients in the latter were not hormone refractory. Median follow-up after initial MRI was 11 (range 1–50) months. Patients with rSCC received radiotherapy to the site of the lesion and some received bisphosphonate treatment as indicated by the treating physician.

Kaplan–Meier analysis was used to investigate neurological deficit-free survival (NDFS) (i.e. the time to development of neurological deficit); those patients who did not develop neurological deficit were censored at time of death or at end of follow-up. Cox's multivariate regression was used to identify influential risk factors for neurological deficit. The following variables were tested: rSCC during first MRI; PSA level at the time of initial MRI; PSA doubling time; radiotherapy; and back pain.

Results

Median overall survival was 416 (95% CI 23 to 987) days.

The median time to development of neurological deficit was 896 (95% CI 13 to 986) days (Figure 5). Thirty-seven (28.5%) of the 130 patients had rSCC at initial MRI. When patients were dichotomised into those with (n = 37) or without (n = 93) rSCC at the first MRI, the latter exhibited slower development of deficit (Figure 6) but this did not reach statistical significance (p = 0.103 by log-rank test); median times were 657 (95% CI 23 to 1103) days and 896 (95% CI 13 to 986) days, respectively.

FIGURE 5.

Time to development of neurological deficit after initial MRI (patients' initial MRI requested because of physicians' judgement of a high risk of SCC). Patients were censored either at the time of death or at the time of last follow-up for surviving patients who had not developed neurological deficit. Data read from published graph (Venkitaraman 2010133).

FIGURE 6.

Time to development of neurological deficit according to rSCC at initial MRI. Patients were censored either at the time of death or at the time of last follow-up for surviving patients who had not developed neurological deficit. Data read from published graph (Venkitaraman 2010133).

Of the 37 patients who had rSCC during initial MRI, 10 developed a repeat rSCC on MRI follow-up after a median time of 161 (95% CI 63 to 259) days. In 6 out of 10 of these patients, recurrence was at the same site as initial rSCC and radiotherapy. Six of the 37 patients (16.2%) developed irreversible paraparesis on follow-up.

Of the 93 patients without rSCC at initial MRI, 20 (21.5%) developed SCC during follow-up. The median time to development of an rSCC for patients with no rSCC on initial MRI was 283 (95% CI 229 to 337) days, and 8 out of 93 (8.6%) developed paraparesis during follow-up.

High PSA level at the time of initial MRI (HR 2.04, 95% CI 1.05 to 3.96; p = 0.035) and short PSA doubling time (< 3 months) (HR 0.40, 95% CI 0.19 to 0.79; p = 0.009) were found to significantly predict adverse NDFS on univariate analysis, but rSCC on initial MRI (p = 0.11), radiotherapy (p = 0.1) and back pain (p = 0.059) did not attain statistical significance. On multivariate analysis, only a rapid PSA doubling time (< 3 months) independently predicted future neurological deficit (p = 0.042).

The authors tabulated the actuarial NDFS at several time points for patients with and without back pain and for patients with and without rSCC at initial MRI. These data are summarised graphically in Figure 7.

FIGURE 7.

Actuarial deficit-free survival according to (a) back pain and (b) rSCC at initial MRI.

Author conclusions

Imaging by MRI of the spine can be used to detect asymptomatic rSCC in patients with castration-resistant prostate cancer. Serial estimations are required to maintain a low incidence of clinical SCC. If serial MRI is to be used to detect rSCC in 90% of patients before the development of neurological signs, the optimum frequency for scanning depends on the subset of patients studied.

Reviewer conclusions

The study findings are consistent with the fairly obvious conclusion that, among castration-resistant prostate cancer patients lacking neurological deficit, those in whom MRI is suggestive of occult SCC (i.e. with rSCC) will develop neurological deficit sooner than those patients whose MRI scan is negative for occult SCC. Only 37 (28%) of patients had occult SCC so the study lacked power. Rapid escalation of serum PSA was found to be associated with increased risk of neurological deficit.

Relevant aim

The aim of this study was to determine the correlation of specific symptoms and signs with epidural spinal cord compression (ESCC) in patients suspected of having ESCC, and to compare the roles of skeletal radiographs, radionuclide scans and CSF analyses in the diagnosis of SCC.

Design and method

This was a retrospective study based on case records from a single centre. Consecutive patients with a positive diagnosis of breast cancer from records dated between January 1977 and July 1982 were selected if they had received myelography for suspected SCC secondary to metastatic breast cancer. The number with suspected SCC who did not receive myelography was not reported.

Results

In 42 of the 78 included patients (age range 22–75 years), myelography was positive for ESCC (with complete block in 21); myelography was negative in 36 patients. In all 42 myelogram-positive patients, bone scans were also positive for spinal metastases, and most had three or more of these at different sites. Bone scans were positive for spinal metastases in 11 of the 36 patients with negative myelograms. Seven of the patients with a negative myelogram were found not to have metastatic disease.

There was no statistical difference between myelogram-positive and myelogram-negative groups in the distribution of spinal metastases as detected by bone scan or in the proportion with visceral metastases (Table 14). The distribution of spinal metastases in patients not undergoing myelography is unknown. The distribution of various signs and symptoms of SCC between patients with positive and negative myelograms indicated that back pain, paraesthesias and bladder/bowel dysfunction were more common in the myelogram-positive group (p≤ 0.05 by chi-squared test; Table 15). The number of metastatic sites (0, 1, 2, ≥ 3) and the type of dominant site (bone, visceral, soft tissue, none) were also significantly associated with a positive myelogram. Plain radiographs were analysed in only 13 of the 36 patients with negative myelograms and therefore comparison between groups based on radiographic findings is unlikely to be meaningful.

Author conclusion

Specific signs and symptoms may predict SCC. Back pain, radicular pain, paraesthesias and even bladder and bowel dysfunction are seen in patients without cord lesions. Myelography remains the most precise tool for diagnosing spinal cord lesions.

Reviewer conclusions

In a comparison of patients in whom myelography for suspected ESCC was positive or negative, chi-squared tests indicated that a positive bone scan, back pain, paraesthesia and bladder/bowel dysfunction at the time of myelography were more common in patients with a positive myelogram than those with a negative myelogram. The discriminatory power of these signs or symptoms to distinguish patients with or without ESCC was poor because they were common in both groups.

Relevant aim

The aim of this study was to examine potential clinical neurological and oncological risk factors for CT-established SCC in breast cancer patients with suspected SCC.

Design and method

This was a retrospective study of 123 episodes of suspected SCC encountered at a single centre over a 2.5-year period from 1985 to 1988. In all, 405 episodes of suspected SCC were recorded; the number of breast cancer patients among these was not reported. Sixty-three episodes were excluded for various reasons including radiotherapy to the suspected site of SCC within 1 year of the index CT scan. Of the remaining 342 episodes, 146 occurred in breast cancer patients; 23 episodes were excluded because of incomplete medical records. The 123 included episodes involved 93 patients with a median age of 52.9 (range 29.8–77.3) years, of whom 98% had previously established metastatic disease and 89% had skeletal involvement. The spinal level of the suspected episodes was reported as lumbar 38%, thoracic 35%, cervical 18%, sacral 1% and various combinations (8%).

All patients received CT scans; problematic cases were examined with MRI when this became available. Bone scans were performed in 99% of episodes and plain spinal radiography was undertaken in 72%.

Univariate analysis (Fisher's exact test) was used to test for association of potential clinical, neurological and oncological factors predictive of SCC. Stepwise multivariable logistic regression was used to identify independent risk factors for SCC and Kaplan–Meier analysis examined survival of SCC-positive and -negative patients after the index CT scan.

Results

Of 123 episodes (93 patients), the index CT scan revealed spinal cord (or cauda equina) displacement in 14 (11%), thecal sac compression (TSC) without displacement in 19 (16%), epidural cancer without displacement or compression in 21 (17%) and epidural cancer in 69 (56%). Depending on the definition of clinically important metastatic ESCC, the authors estimate that the CT scan was positive in between 11% and 44% of episodes (any epidural disease). In their analysis of predictors, the authors took clinically important metastatic ESCC to be TSC or thecal compression (equivalent to 33 of 123 episodes, 27%).

In univariate analysis, predictors for a positive CT index scan for clinically important metastatic ESCC were: known bone or vertebral metastases for ≥ 1 year; metastatic breast cancer at initial diagnosis; previous spine radiotherapy; objective weakness; increased deep tendon reflexes; and abnormal plantar reflex. These were weak predictors; the best was objective weakness with specificity of only 67% and positive predictive value of 40%. Given these data it is possible to populate a 2 × 2 table for this predictor, as shown in Table 15.

Some variables tested occurred so infrequently in the population that they could not reach statistical significance as predictors.

In multiple logistic regression analysis, four independent predictors of TSC were identified as follows: known bone metastases ≥ 2 years (OR 3.0, 95% CI 1.2 to 7.6; p = 0.02); metastatic disease at initial diagnosis (OR 3.4, 95% CI 1.0 to 11.4; p = 0.05); objective weakness (OR 3.8, 95% CI 1.5 to 9.5; p = 0.005); and vertebral compression fracture on spine radiograph (OR 2.6, 95% CI 1.0 to 6.5; p = 0.05). These four predictors stratified episodes into subgroups with widely varying risk of positive TSC scans, ranging from 12% (no risk factors) to 85% (≥ 3 risk factors) as shown in Table 16. Of the 123 episodes, 11% were associated with three or more risk factors relative to a prevalence of 27% for TSC.

Author conclusions

The results suggest that evaluation of breast cancer patients with suspected SCC might include clinical information about disease course, in addition to neurological examination and previous imaging studies. If confirmed, these predictors may help clinicians to assess risk in this patient population.

Reviewer conclusions

A combination of three or four of the four identified risk factors predicted TSC with a probability of 85% in this population of breast cancer patients. In the current context the presence of any two of these would probably lead to physicians requesting MRI examination, and so their predictive utility would appear to be limited as these patients are obvious cases for further imaging examination. The study by Talcott et al. 131 examined CT scans for suspected SCC over the same period (1985–8) at the same institute; Talcott et al. included several primary cancers, of which breast cancer represented 42% of 258 patients (n = 108); it appears that data for most or all of the patients in Lu et al. 120 were also included in the Talcott study. 131

Relevant aim

The aim was to develop a radiological method to assess vertebral deformity in women and to employ the method to estimate the incidence and prevalence of vertebral deformity in a population of women with breast cancer.

Design and method

This study used radiographs of the spinal vertebrae of 100 randomly selected normal women (aged 45–50 years) to measure the anterior, posterior and central heights of vertebrae T4 to L5; these measures were termed A, P and C. The predicted posterior height (PP) of a given vertebra was estimated using the mean p-values for adjacent vertebrae (for the population of 100 women). The predicted height for each vertebra (for 100 women) could then be compared with its measured height (for 100 women) to give a ratio and its SD. For the posterior height, this was termed the posterior to predicted posterior height (P/PP) ratio. All P/PP ratios were very close to unity (listed as 1.000) and SDs were mostly in the range 0.04–0.06 and varied somewhat between each specific vertebra. This indicated that estimating the predicted height of a vertebra from the heights of normal adjacent vertebrae was a reasonably accurate procedure. The ratios A/P and C/P and their SDs were also calculated. The authors tested the within- and between-observer reproducibility of the procedure and judged that it was satisfactory.

The ratios P/PP, A/P and C/P were plotted in a normal probability plot and each was shown to have a normal distribution. Given these ratios and their normal distributions, the authors defined criteria for the presence of deformity in a given vertebra in terms of how many SDs the ratios for a scrutinised vertebra departed from the mean observed among the 100 normal women (taking the SD specific for that vertebra as determined for the 100 normal women). The number of standard variations required (e.g. 2.5, 3, 3.5, 4, 4.5) could be varied in determining the presence or absence of deformity and so various cut-off thresholds for deformity could be investigated. Criteria for four types of deformity (central collapse, anterior wedge, posterior wedge and crush deformity) were defined in terms of the combination of ratios required to be satisfied according to appropriate SDs for different cut-offs.

An estimate of the specificity of a particular cut-off threshold was made by examining the vertebrae of the 100 normal women and assuming a prevalence of zero and that a vertebra below the mean according to the defined threshold represented a false-positive (Figure 8 and cell B in Table 17). This allowed the population of cells B and D in the 2 × 2 table and thence the calculation of specificity.

FIGURE 8.

Illustration of the proportion of false-positives among normal vertebrae at a defined threshold.

Vertebrae of 163 women with breast cancer (aged 30–75 years) and with skeletal metastases were examined and classified according to the presence of vertebral deformity using A/P, C/P, P/PP ratio criteria and five SD cut-offs ranging from 2.5 to 4.5. After 6 months the vertebrae were re-examined in 121 of these women and newly developed deformities were estimated. These evaluations allowed estimates of point prevalence and 6-month incidence rates of vertebral deformity for this population. For the breast cancer population false-positives (cell B in Table 17) were defined as those with ratios above the mean (for normal women) by a defined number of SDs (depending on the threshold cut-off).

Results

For normal vertebrae the P/PP ratio was normally distributed with a mean of 1.00; SDs of the P/PP ratio varied from vertebra to vertebra, suggesting the need to use site-level criteria in determining the presence of deformity. The SDs for the P/PP ratio were similar whether the PP was made using measures from one or from four adjacent vertebrae (un-deformed). Using a cut-off of 3 standard deviations, the prevalence of vertebral deformity in women with breast cancer and skeletal metastases was 46%.

Author conclusions

The technique developed for assessment of vertebral deformities is robust and rapid, and has minimal effects on sensitivity while maximising specificity. The method was able to detect minor vertebral deformities, which subsequently progress, and there is a close relationship between existence of deformities and subsequent rate of deformity in breast cancer.

Reviewer conclusions

Radiography coupled with vertebral measurements and the use of the criteria developed by the authors allowed specific detection of vertebral deformity in women with breast cancer and skeletal metastases. Such detection before the development of frank neurological involvement could be useful. Radiography of the spine is not now used in the comprehensive way reported in this study and whether the procedures developed could be applied using CT or MRI images is uncertain; however, the loss of ‘length’ dimension in a diseased vertebra, relative to the value of that dimension expected from measurement of healthy adjacent vertebrae, is currently used to determine if fracture is present.

Relevant aim

This study attempted to identify prognostic factors for bone metastases.

Design and method

The main focus of this study was to provide basic data on the incidence of bone and spinal metastases and SCC of Japanese breast cancer patients treated with endocrine therapy or chemotherapy following primary surgery in a single institution and to calculate the survival rate after breast surgery, bone or spinal metastasis, and paralysis due to cord compression, using the Kaplan–Meier analysis.

This was a retrospective study of 695 patients with breast cancer (four men, 691 women) who underwent radical surgery at a single centre between January 1990 and December 1996; mean age at surgery was 53.1 (range 24–88) years. Forty-two patients had other concurrent cancer. One hundred and three patients had visceral metastases at baseline and 592 had no visceral metastases.

Various imaging methods were employed including bone scintigraphy, chest radiography, chest CT, liver ultrasonography, abdominal CT, cranial CT and MRI (or any combination of these). A Cox proportional hazards model was used to identify prognostic factors for the development of skeletal metastases.

Results

Bone metastases were observed in 148 patients at the end of the observation period (all received chemotherapy; 44 patients received endocrine therapy before the metastases). The survival of these patients after surgery was much worse than that of the 547 patients who did not develop skeletal metastases (5-year survival 45.8% vs. 89.9%). The interval between surgical treatment and the development of bone metastases ranged from 0 to 130 months (median 19 months). Kaplan–Meier analysis indicated that after surgical treatment of breast cancers, bone metastases developed in 18.1% of the patients over 5 years and in 24.7% of the patients over 10 years (Figure 9).

FIGURE 9.

Time to development of skeletal metastases after surgery for breast cancer. Data read from published graph (Oka 2006123); Weibull distribution fitted to the first 10.5 years of data. Number at risk at t0 assumed to be 605.

The risk factors for development of bone metastases identified using Cox's regression were tumour stages evaluated by TNM classification (HR 1.615, 95% CI 1.322 to 1.973; p < 0.0001); nodal (N) stage classification (HR 2.128, 95% CI 1.381 to 3.279; p = 0.0006); the presence or absence of metastases to axillary lymph nodes (p = 0.0006); and the presence or absence of metastases to important organs (HR 7.502, 95% CI 5.100 to 11.036; p < 0.0001). By the end of the observation period spinal metastases were observed in 121 of the 148 patients who developed skeletal metastases, and 17 out of 121 of these developed paralysis due to SCC. The time between detection of skeletal metastases and development of SCC ranged from 2 to 72 months with a median of 4.4 months.

The study investigated factors prognostic for survival after surgery and survival after development of skeletal metastases. Postsurgery prognostic factors included tumour stages evaluated by TNM classification (HR 1.346, 95% CI 1.099 to 1.648; p = 0.004); N stage classification (HR 1.524, 95% CI 1.030 to 2.257; p = 0.03); the presence or absence of metastases to axillary lymph nodes (p = 0.03); presence or absence of metastases to important organs (HR 3.356, 95% CI 2.226 to 5.060; p < 0.0001); presence or absence of oestrogen receptors (HR 1.686, 95% CI 1.102 to 2.580; p = 0.02); presence or absence of progesterone receptors (HR 1.954, 95% CI 1.274 to 2.997; p = 0.002); and the presence or absence of bone metastases (HR 3.704, 95% CI 2.415 to 5.682; p < 0.0001).

Prognostic factors for survival after development of bone metastases included the presence or absence of metastases to important organs (HR 2.379, 95% CI 1.484 to 3.815; p = 0.0003) and the presence or absence of progesterone receptors (HR 2.689, 95% CI 1.553 to 4.657; p = 0.0004). Statistically, there were no factors significantly associated with the prognosis of breast cancer patients with paralysis due to cord compression (only 17 patients available for analysis).

Author conclusions

To detect predictive factors of long survival after paralysis and establish indications for surgery, a comparative study among large groups of patients with paralysis and with different backgrounds is needed.

Reviewer conclusions

Risk factors for the development of skeletal metastases were tumour stage (TNM classification), N stage classification, metastases to axillary lymph nodes and visceral metastases. The prognostic factors for survival after development of bone metastases were visceral metastases and presence of progesterone receptors.

Relevant aim

The aim was to identify factors that predict complications from skeletal disease in patients with bone metastases from advanced breast cancer.

This was a retrospective study of breast cancer patients with bone metastases identified between 1975 and 1991 at one centre. Of 1437 patients so identified, 111 were followed up elsewhere and 460 were diagnosed elsewhere and their records contained too little information for inclusion in the study. The notes for 7 patients (0.5%) were not found. The remaining 859 patients were included. Patients were divided into four groups according to sites of disease at the time of diagnosis of bone metastases as follows: (1) bone metastases only (n = 243, 28%); (2) bone and soft tissue disease only (n = 268, 31%); (3) bone and pleuropulmonary disease, with or without soft tissue disease (n = 237, 28%); (4) bone and liver metastases, with or without soft tissue or pleuropulmonary disease (n = 111, 13%). Patients were monitored with scintigraphy and radiography; myelography was not mentioned.

Results

The time from diagnosis of skeletal metastases to vertebral fracture was shortest in the bone-only group (p < 0.0017). Figure 10 illustrates the shape of the reported Kaplan–Meier plots for each group. In addition, patients with bone-only disease developed SCC more rapidly than patients in other groups (p = 0.01). Thirty-six patients with bone-only disease at diagnosis of bone metastases (15%) developed cord compression compared with 2–6% of patients in the other groups. Bone scan evidence of metastases in the spine did not predict for subsequent development of cord compression.

FIGURE 10.

Time to development of skeletal fracture according to patient's subgroup.

Survival from diagnosis of bone metastases was significantly greater for patients with bone disease only at diagnosis of skeletal metastases (p < 0.001), and was shortest for patients with concomitant liver metastases (median survival 5.5 months).

There were no differences between the groups in the time to pathological long bone fractures. However, since patients with bone disease only at diagnosis of skeletal disease lived longest, most fractures occurred in this group. Of a total of 243 such patients, 42 (17%) developed a pathological long bone fracture (i.e. 1 in 5.8 patients), compared with 5 of 111 shorter-lived patients with bone and liver disease (5%) (i.e. 1 in 22.2 patients).

Author conclusions

The results suggest that patients with disease confined to the skeleton at the diagnosis of bone metastases are most likely to develop skeletal-related complications from advanced breast cancer. Such patients may benefit most from treatment with bisphosphonates.

Reviewer conclusions

The study does not give detailed information regarding participants such as age or time since diagnosis. In breast cancer patients diagnosed with bone metastases, longer survival is a risk factor for vertebral fracture and for SCC, and longer survival is associated with lack of disease at sites additional to the skeleton, especially liver disease and to a lesser extent pleuropulmonary disease.

Relevant aim

The aim was to investigate if prognostic factors identified ex vivo using structural rigidity analysis of transaxial CT image data predict in vivo vertebral fracture in cancer patients with spinal metastases more effectively than contemporary clinical and radiographic guidelines published by Taneichi et al. 89

Design and method

This study describes the development of the authors' biomechanical model for prediction of fracture. The model estimates the load-bearing capacity of vertebrae using transaxial CT scans. It was tested in 106 women with breast cancer metastatic to the spine who were followed up for 4 months after CT scan to monitor vertebral fracture (4 months was selected as a sufficiently short period for the influence of any tumour progression that might occur to minimally affect risk).

The fracture risk index (FRI) was calculated for each vertebra between T8 and L5 using two different load scenarios for each patient: (1) lifting a 10-kg mass and (2) rising from a chair. A FRI > 1 predicts that fracture would occur during the applied load condition. The accuracy of FRI was compared with the radiographic criteria according to the Taneichi et al. guidelines,89 which predict fracture on the basis of size and location estimates of the spinal tumour. Investigators blinded to the predictions used MRI and radiography to establish the incidence of actual fractures using standard criteria.

Results

Over the 4-month period, 10 out of 106 patients suffered one or more new vertebral fractures. Both the CT-based structural rigidity analysis and the Taneichi criteria predicted that these 10 patients were at increased fracture risk (sensitivity = 100% for either method, threshold set so there are no false-negatives).

However, the CT rigidity analysis was better at predicting which patients would not fracture an affected vertebra (specificity = 49% when FRI > 1 for lifting a 10-kg mass) compared with the Taneichi CT criteria (specificity = 20%). With sensitivity at 100%, negative predictive ratios are not calculable (infinitely large). The calculated positive LRs were modest for both methods (1.96 and 1.25, respectively), as also were the positive predictive ratios (0.17 and 0.11).

When the load-carrying capacity of the vertebra was normalised by the patient's body mass index (BMI; kg/m²) and the threshold for predicting vertebral fracture set to achieve 100% sensitivity, the specificity for predicting no vertebral fracture was improved to 69% (i.e. a false-positive rate of 31%). This is illustrated in purely diagrammatic form in Figure 11. The negative LR becomes 0.25 and positive LR 3.2.

FIGURE 11.

Diagrammatic representation of threshold set to 100% sensitivity and 69% specificity. The dashed vertical line represents the load capacity threshold for discriminating predicted fracture and non-fracture.

By logistic regression the estimated RR for fracture based on FRI > 1 was 4.2 (95% CI 1.4 to 12.8; p < 0.001). When controlling for BMI (kg/m²), the adjusted RR for fracture based on FRI > 1 was 7.9 (95% CI 1.8 to 34.5; p < 0.001).

Author conclusions

A CT-based structural rigidity analysis was as sensitive but significantly more specific than the best radiographic guidelines for estimating metastatic cancer vertebral fracture risk.

Reviewer conclusions

The paper has inadequate information in terms of patient population. The number of events (10 patients with fractures) was small. The study compares sensitivity and specificity of CT-based structural rigidity analysis against available guidelines (Taneichi et al. 89); neither method performed well in this population.

Relevant aim

These authors aimed to compare CT-based structural rigidity analysis with current standards for prediction of spinal fracture in women with breast cancer with spinal metastases. The current standard is implied to be plain radiography used with an empirically derived logistic regression analysis based on size and location of vertebral metastases observed by axial CT scanning (Taneichi's algorithm).

Design and method

The records of 1024 breast cancer patients at a single institute were reviewed to identify study participants. Ninety-four patients were included; patients were excluded if records indicated neural compromise (due to metastases in brain or spinal cord), withdrawal, relocation, death, previous fracture at metastatic or adjacent site, surgical treatment for impending fracture or fractured bones due to significant trauma. Fifty-one per cent of included patients were postmenopausal, many with co-existent osteopenia. It was unclear when CT and radiographic examinations were conducted relative to time of selection into the study.

Axial CT scans were used to estimate rigidity, a product of bone tissue modulus and geometry. It had been previously established (in an ex vivo study) that the force needed to fracture vertebrae is proportional to the weakest cross-section through the affected bone; thus, the scans were used to identify the cross-sectional structural rigidity with weakest resistance to axial (EA), or bending (EI) loads (that is the minimal axial load and bending load rigidities for each vertebra). From this, the load-bearing capacity (LBC) of the vertebra in combined axial compression and forward bending was also estimated using ‘beam theory’. The LBC was standardised on BMI (kg/m²) (LBC/BMI). The rate of fractures over the next 4 months was recorded by independent investigators.

Results

The value for each of the four parameters (EI, EA, LBC, LBC/BMI) in each of the 247 vertebrae was estimated. There were 11 fractures over 4 months (236 vertebrae did not fracture). Fractures were distributed as shown in Table 18.

The value for each of the four parameters in each of the 247 observed vertebrae was calculated. From these values, for each parameter the maximum value for the 11 fractured vertebrae was selected as the diagnostic threshold for that parameter. For example, for LBC/BMI, the maximum value among fractured vertebrae was 46.5. As all other fractured vertebrae had values < 46.5, using this as the threshold meant that all fractures would be detected; the resulting sensitivity was 100%. Of the 236 unfractured vertebrae, 74 also had a LBC/BMI < 46.5, giving a specificity of 68.6% [(236 − 74)/236, which was reported as 70%, possibly based on patients rather than vertebrae]. Should a successful treatment be available that prevented fracture, these results imply a number needed to treat of ≈ 7.7.

Using the same procedure for LBC, EI and EA, the specificities were 44%, 53% and 55% (all sensitivities at 100%), respectively. Using Taneichi's algorithm and sensitivity set at 100%, specificity was only 20% (i.e. very many false-positives).

The authors provided a receiver operating characteristic (ROC) curve for LBC/BMI showing how sensitivity and specificity were affected by changing (reducing) the value of the cut-off (Figure 12). Hence, as cut-off fell below 46.5, < 100% of the fractures were detected, but there were fewer false-positives and so specificity improved. The area under the ROC curves was estimated using a binomial semi-parametric model. The results were Taneichi, 0.6; LBC, 0.82; EI, 0.80; EA, 0.68; LBC/BMI, 0.84. Corresponding p-values for the comparison with chance (tossing a coin; area under the curve = 0.5) were 0.25, 0.001, 0.001, 0.002 and < 0.001, respectively.

FIGURE 12.

Receiver operating characteristic curves for LBC/BMI and radiological prediction. Data read from published graph (Synder 2009128).

Author conclusions

The CT-based structural rigidity analysis is as sensitive as, and significantly more specific than, current radiographic criteria for predicting vertebral fracture in breast cancer.

Reviewer conclusions

Patient selection was not fully described; it is possible that sensitivities and specificities could vary depending on stage of vertebral invasion by metastases, so selection of participants is important. From T8 to L5 for 94 women provides at least 658 potential vertebrae examined; 247 were used for parameter calculations, but it was not reported how these were selected (i.e. whether these were all those identified with metastases or a proportion of them). It is unclear if the patients in this study overlapped with those in the Snyder 2005127 study considered above. There were only 11 events (fractures); to retain 100% sensitivity with more events would probably require moving the threshold to a higher value, thereby probably increasing the rate of false-positives and reducing specificity.

Relevant aim

The stated aim was to identify the risk factors for SREs in patients with advanced NSCLC.

Design and method This was a retrospective study of 642 NSCLC patients. According to the report, for inclusion in the study, patients required a histological or cytological diagnosis of NSCLC, stage IV disease or postoperative recurrence with distant metastases, and ‘no prior chemotherapy’ or ‘chemotherapy prescribed by the National Cancer Center Hospital between 2000 and 2006’. These criteria may therefore have excluded patients who received certain sorts of chemotherapy not prescribed by the National Cancer Center Hospital between 2000 and 2006. However, all 642 patients were described as having received first-line chemotherapy as follows: platinum based, n = 429; gefitinib (Iressa^®, AstraZeneca), n = 117; third-generation monotherapy, n = 47; non-platinum doublets, n = 9. Patients were excluded if they had postoperative local recurrence without distant metastases. Forty-three patients received zoledronic acid (Aclasta^®; Novartis) either before (n = 26) or after (n = 17) the development of SREs.

At initial diagnosis 399 had no bone metastases, 63 had a single bone metastasis and 180 had multiple bone metastases. Disease progression was observed in 580 out of 642 patients; the initial site of progression was bone in 78 and other than bone in 502.

SREs were defined as (1) pathological fractures, (2) SCC, (3) requirement for radiation therapy, (4) requirement for surgery to the bone, (5) requirement for radiological intervention to the bone, and (6) hypercalcaemia of malignancy that was either fatal or required emergency treatment. Association of baseline characteristics with development of SREs was examined in univariate analysis and multivariate logistic regression. Cox's proportional hazards model was used to identify risk factors for time to event. Kaplan–Meier analysis was used to investigate the time to first SRE after commencement of chemotherapy.

Results

A total of 118 (18.4%) patients developed SREs during or after initial chemotherapy (107 required radiotherapy to bone, five developed hypercalcaemia of malignancy, three developed compression fracture of vertebrae, two required surgical treatment of the bone and one underwent radiofrequency ablation therapy to bone).

In univariate analysis the number of bone metastases (none, single or multiple) at initial diagnosis (p < 0.001) and history of radiotherapy to bone before chemotherapy (p = 0.001) were associated with the development of SREs, whereas sex, age, performance status and cancer histology were not (Table 19). However, multivariate analysis using a logistic regression model showed that the number of bone metastases was strongly associated with the occurrence of SREs (OR 3.08, 95% CI 1.60 to 5.94, for single bone metastasis; OR 4.27, 95% CI 2.66 to 6.86, for multiple bone metastases), whereas radiotherapy to the bone before the chemotherapy was not (OR 1.43, 95% CI 0.69 to 2.97).

Of patients who had no bone metastasis at diagnosis, only 10.3% developed SREs, whereas 27% of patients with a single bone metastasis and 33% of patients with multiple bone metastases developed SREs during their clinical course (p < 0.001); the median follow-up for SREs was 10.4 (range 0.1–77) months.

In univariate analysis the time to development of SREs was associated with sex, performance status, number of bone metastases at diagnosis and history of radiotherapy to bone before chemotherapy. However, on multivariate analysis a history of radiotherapy to bone was not associated with the development of SREs (Table 20).

Kaplan–Meier analysis indicated that the time to first SRE was shorter for those with multiple bone metastases at diagnosis than those with a single metastasis or none; the relationship published is illustrated in Figure 13.

FIGURE 13.

Time from start of chemotherapy to first skeletal event according to bone metastases at diagnosis. Data read from published graph (Sekine 2009125). Patients with no event by end of follow-up were censored at that time.

For the analysis of SRE-free survival, a SRE or death was taken as an event and patients without an event by end of follow-up were censored. In multivariate analysis, SRE-free survival was strongly associated with performance status (compared with zero performance status: HR 1.47, 95% CI 1.15 to 1.89, for performance status 1; and OR 3.72, 95% CI 2.31 to 5.98, for performance status 2 or 3). The median SRE-free survival was 23.5 months (95% CI 18.6 to 28.5 months) in patients with performance status of 0, 13.1 months (95% CI 10.4 to 15.8 months) in patients with performance status of 1 and 5.2 months (95% CI 1.0 to 9.4 months) in patients with performance status of 2 or 3 (p < 0.001). To a lesser extent male sex and multiple bone metastases at diagnosis were also indicators of poor SRE-free survival.

Author conclusions

The presence of multiple bone metastases was significantly associated with the development of SRE in patients with advanced NSCLC treated by systemic chemotherapy. The factor ‘multiple bone metastases’ was identified as a risk factor for the development of SREs as assessed by all three parameters, and was, therefore, considered as a definite risk factor for the development of SREs. Male sex and poor performance status may be additional risk factors for the development of SREs in these patients. Male sex and poor performance status were significant risk factors influencing the SRE-free survival, marginally significant in relation to the time to the first SRE, and not significant in relation to the presence of SRE.

Reviewer conclusions

This was a large study with homogeneous mixed cases and more statistical power than many others. The definition of SRE included a number of clinical presentations so it is difficult to distinguish the number of occurrences related to spines. The study does not report number of spinal metastases. A small proportion of participants used bisphosphonates, a drug that prevents loss of bone mass/delayed SREs. It is probably not a surprising finding that the greater the number of bone metastases, the greater the risk of a SRE.

Relevant aim

The study aimed to identify clinical factors that can predict SREs in patients with advanced NSCLC.

Design and method

Patients were identified from medical records of consecutive diagnoses of advanced NSCLC at a single centre between January 2006 and March 2009 (n = 1166). From these, 273 patients with bone metastases secondary to NSCLC were identified from imaging (e.g. scintigraphy and PET) and biopsy records. Clinical data were obtained from the date of primary diagnosis to 31 October 2009; median follow-up was 11 (range 0.7–46.0) months. Of the 273 included patients, 242 were diagnosed with bone metastases at the time of NSCLC diagnosis. Bone metastases were found at multiple locations in most patients. A total of 528 locations were identified (221 to the spine).

The authors investigated the following potential risk factors for their association with SREs: sex; ever a smoker; adenocarcinoma/non-adenocarcinoma; no history of therapy with an epidermal growth factor receptor tyrosine kinase inhibitor (EGFR TKI) such as gefitinib; Eastern Cooperative Oncology Group (ECOG) status; BMI (kg/m²); and age. These were evaluated in univariate analysis and multivariate logistic regression for an association with the risk of experiencing a first SRE (i.e. at least one event), for the time to first SRE using Kaplan–Meier methods, and for the risk of recurrent SREs (> 21 days after the preceding event) using survival-adjusted multiple event analysis. 137

Results

Of the 273 patients analysed, 171 experienced at least one SRE, and 46 had multiple SREs. A total of 229 SREs developed of which 65 occurred before any systemic treatment was received. The most frequent site of SREs was the spine (55.2%). The pattern of SREs was complex: radiotherapy in 169 cases (73.9%), cord compression with vertebral fracture in 14 cases (6.1%), cord compression without definitive vertebral fracture in 14 cases (6.1%), pathological fracture in 30 cases (13.1%), and one case each of prophylactic surgery for impending fracture and hypercalcaemia (0.8%).

In multivariate analysis, only ‘ever smoked’ was associated with significantly higher SRE risk (OR 2.8, 95% CI 1.32 to 6.00). The same result was obtained if patients who received bisphosphonate therapy were omitted from the analysis. For median time to first SRE in multivariate analysis (Table 21) the following variables were associated with shorter median time to first SRE: no history of therapy with a EGFR TKI such as gefitinib; ever smoked; and histology of non-adenocarcinoma.

The same three factors (no history of therapy with a EGFR TKI, ever smoked and histology of non-adenocarcinoma) and also ECOG status 2/3 were significantly associated with increased risk of multiple events (separated by at least 21 days) (Table 22).

Significantly more SREs per cycle of treatment occurred during cytotoxic therapy than during EGFR TKI therapy; however, the authors draw attention to the potential pitfall for interpretation in that systemic therapies did not necessarily precede SRE in all cases.

Author conclusions

This study suggests that metastatic NSCLC patients with characteristics such as ever having smoked, no history of EGFR TKI therapy, poor ECOG status and non-adenocarcinoma are more likely to suffer SREs.

Reviewer conclusions

SREs appear to have been classified as pathological fracture, SCC with or without vertebral fracture, need for radiation or surgery to bone and hypercalcaemia of malignancy. The risk factors identified may well apply equally to SCC and/or vertebral fracture alone; however, this would need to be investigated using the appropriate narrower definition of an event. This is one of the few studies that considered the risk of repeated events.

Relevant aim

The aim was to undertake an analysis of medical records to define factors predictive of SCC.

Design and method

This was a retrospective analysis of medical records of patients who participated in an RCT that investigated the effectiveness of chemotherapy regimens for SCLC between 1982 and 1986. In this trial, participants (n = 616) were randomised to four or eight courses of vincristine (Oncovin^®, Genus), cyclophosphamide (trade names Endoxan, Cytoxan, Neosar, Procytox, Revimmune) and etoposide [etoposide phosphate or VP-16 (current brand name Etopophos)]. On relapse, patients were further randomised to standard care or to further therapy with doxorubicin INN (trade name Adriamycin; also known as hydroxydaunorubicin) and methotrexate. The cases of SCC (n = 24, 3.8% of 616 patients; age range 42–64 years; 20 male) consisted of those trial participants who according to medical records had a diagnosis of SCC at the start of the trial or who developed SCC during follow-up. SCC was assessed on clinical grounds of signs and symptoms. Of the 24 with SCC, myelography was performed in only 11.

Results

Of 616 patients, 500 had undergone bone scanning at presentation (presumed to be study entry). Among these, scanning was judged to be positive for spinal metastases in 131 patients; 9 of these 131 patients (6.9%) had a diagnosis of SCC at some time. Hence, 15 patients (the 24 patients with SCC minus the 9) had SCC at some time but either did not have a scan or had a negative scan. If all 15 had negative scans for spinal metastases then the percentage of any-time SCC patients who had negative scans at presentation is 4.1%. This is somewhat lower than the 6.9% SCC associated with positive scans for spinal metastases. If the 15 SCC patients not detected with a positive spinal bone scan were proportionately distributed among all non-positive scan patients (616 − 131 = 485) then percentage of any-time SCC patients providing negative scans for spinal metastases reduces to 3.1%, half that for positive scans indicating a likely association of spinal metastases with present or future SCC. Two of the patients who did not have bone scans had plain radiographs showing vertebral collapse.

Of 24 patients (3.9% of 616) who presented with back pain and a positive spinal bone scan, nine (38%) had SCC at some time. Of 32 patients presenting with cerebral metastases, 12.5% had SCC at some time. Of 87 patients who relapsed with cerebral metastases, 8% had SCC at some time.

Author conclusion

Patients with the combination of cerebral metastases and a positive bone scan had a 25% chance of developing SCC. It may be possible to select patients who should receive radiotherapy to the spine to try to prevent the development of this complication.

Reviewer conclusions

This is an early study with only 24 cases of SCC. Not all SCC cases were confirmed by myelography and the study may have predated wide use of CT or MRI. Multiple logistic regression was not performed and chemotherapy (some patients received very heavy loads of cytotoxic agents), a potentially influential confounder for risk of SCC, was not considered; subsequent studies have indicated that such treatments might affect frequency of SCC. The authors' conclusion regarding the combination of positive bone scan and cerebral metastases as a discriminatory risk factor should be viewed with caution. First, as cerebral metastases and SCC could occur at any time during follow-up, there is no assurance that cerebral metastases preceded SCC. Second, as shown in the 2 × 2 table below (Table 23), viewed as a diagnostic test for SCC, the combination (positive bone scan for spinal metastases plus cerebral metastases) has a sensitivity of only 25% based on the reported results. The positive predictive value is 0.25. The positive LR, that is the ratio of those with SCC to those without SCC returning a positive test, is 0.25/0.03 = 8.33. With a prevalence of 4.1% (24/592) this provides a probability of 0.25 (25%) of having SCC should the test result be positive (pre-test odds = 24/592; post-test odds = (24/592) × 8.33 = 0.338; post-test probability = 1.338/0.338 = 0.252). Patients with a positive test would therefore require further imaging before treatment decisions could be safely undertaken.

Relevant aim

The study aimed to identify risk factors for SCC so as to distinguish between those who would benefit from myelography and those who would not.

Design and method

This retrospective study examined medical records of patients discharged from two hospitals during a period from 1975 to 1980 and identified patients who (1) had clinically suspected epidural compression from ‘metastatic cancer’ and (2) had undergone myelography to confirm or exclude the clinical diagnosis. Myelograms were considered positive for compression if ≥ 80% of the SAS was obliterated. Patient age ranged from 7 to 85 years.

A total of 133 patients were included. They had diagnoses of carcinoma, sarcoma and lymphoma as follows: lung (n = 40), breast (n = 27), prostate (n = 15), lymphoma (n = 12), colon/rectal (n = 6), melanoma (n = 6), kidney and ureter (n = 5), bladder (n = 3), other (n = 15) and unknown (n = 6) (two patients had primaries at two different sites).

Results

The paper reported that 62 and 71 patients, respectively, had positive and negative myelograms. The ratio of positive to negative images varied with primary cancer; for example, 20 of 27 and 3 of 15 myelograms were positive in breast cancer and prostate cancer patients, respectively. Compressions were commonest in the thoracic region, followed by the lumbosacral and cervical regions. Of compressions to the spinal cord, 30 of 47 were complete and 17 were incomplete; there were 15 patients with primarily cauda equina compression.

Stepwise logistic regression was used to examine the association of 13 variables with a positive myelogram. The precise identity of these variables was unclear. The authors selected the following as the most influential variables (p-values were not reported): positive vertebral plain radiographs; sensory level or dermatomal loss on examination; history of local pain; older age; history of weakness; history of radicular pain; male sex; paraparesis or radicular weakness on examination. These predictors were used to calculate the predicted probability of a positive myelogram (compression) for each patient who had compression (i.e. observed compression) and for each who had no compression (observed negative myelogram). Figure 1 of the paper presented the frequency of patients with predicted probabilities of a positive myelogram ranging in 10% steps of probability from 0–0.09 to 0.9–1. Unfortunately, the number of patients with observed positive myelograms shown in the figure fell short of those reported to have positive myelograms (i.e. 62 positive myelograms reported, 53 patients in the figure); this discrepancy was not explained but it is possible the missing patients represent those with incomplete data for logistic regression.

Author conclusion

Attempts to identify symptoms and signs that might increase diagnostic ability were not successful. Logistic regression analysis was used to separate two groups; however, overlap in scores of those with and without compression resulted in difficulty in selecting a useful cut-off point.

Reviewer conclusions

The study sample selected is difficult to define; it will reflect different physicians' decisions about a clinical diagnosis of compression, likely to vary from study centre to study centre and perhaps through time (e.g. 30 years ago vs. now). There was a considerable mix of different cancers in the study sample; any predictive variable(s) uncovered are likely to reflect the mixed cases and would be difficult to generalise to particular conditions or to spinal metastases in general. As different primary tumours manifest at different times, the influence of age as a predictor could relate to mixed cases rather than risk of compression. It is unclear if the type of primary tumour was explored as a predictive variable: p-values were not reported for the logistic regression and there appeared to be errors in the application of the regression results to the study population.

Relevant aim

To understand factors associated with pathological vertebral body compression fractures in patients with metastatic epidural SCC [SCC caused by an epidural mass (EM)].

Design and method

This retrospective study examined medical records of patients who had received surgery for SCC at a single tertiary care centre between 1996 and 2007. Inclusion required MRI evidence of spinal cord displacement by an EM. Patients with more than one discrete lesion, brain metastases, cauda equina or spinal root compression were excluded.

The report implied that, of 216 patients who may have been included, data for 162 were analysed (implying a possibility of ≈ 25% missing data). Patients with vertebral fracture who did not receive surgery were not identified or quantified.

The primary cancer diagnoses among the 162 included patients were various, reported as follows: lung (n = 26, 16%), breast (n = 26, 16%), prostate (n = 20, 12%), renal (n = 21, 13%) and haematopoietic (n = 28, 17%). Other sources included thyroid, gastrointestinal, melanoma and non-renal genitourinary system. The distribution of spinal metastases was described as: cervical, n = 35; thoracic, n = 114; lumbar, n = 49; cervicothoracic, n = 22; and thoracolumbar, n = 24.

Results

Univariate logistic regression identified the following variables that were associated with presurgery vertebral fracture: sensory deficit (p = 0.02), presurgery chemotherapy (p = 0.03), primary breast cancer (p = 0.02), thoracic involvement (p < 0.001), number of spinal levels involved (p = 0.1), number of spinal metastases (p = 0.07) and anterior location (p = 0.005). Variables found not to be associated with vertebral fracture according to univariate regression (p > 0.1) included age, pain symptoms, motor deficit, lytic-type tumour, blastic-type tumour and extraspinal metastases. After multivariate logistic regression, presurgery chemotherapy (OR 2.283, 95% CI 1.064 to 4.898; p = 0.03), primary breast cancer (OR 4.179, 95% CI 1.457 to 11.983; p = 0.008), thoracic involvement (OR 3.505, 95% CI 1.343 to 9.143; p = 0.01) and anterior cord compression (OR 3.213, 95% CI 1.416 to 7.293; p = 0.005) were found to be independently associated with vertebral body compression fractures.

Author conclusion

The factors strongly associated with preoperative compression fractures include lack of sensory deficits, primary breast cancer, anterior spine metastases, thoracic spine involvement, preoperative chemotherapy and possibly preoperative radiation therapy.

Reviewer conclusions

The study sample selected may be unrepresentative of patients with SCC who are at risk of or who have vertebral body compression fractures, as those patients who did not receive surgery were not included; this may or may not be a sufficient proportion to bias results. A further concern is that it appears that 25% of relevant data were missing; however, the report lacks clarity on this. There was a considerable mix of different cancers in the study sample; any predictive variable(s) uncovered are likely to reflect the mixed cases. It would be difficult to generalise to particular conditions or to patients with SCC in general.

Relevant aim

The parts of this study relevant to this report were (1) an investigation of whether tumour type influenced the time from cancer diagnosis to diagnosis of spinal cord or nerve root compression and the clinical severity of the compression; and (2) the identification of prognostic factors for subsequent recurrence of compression at a second site. The authors' main aim focused on the analysis of the prognostic significance of various clinical and radiological variables for ambulatory function and survival following treatment for spinal cord or nerve root compression.

Design and method

This was a prospective study of 153 consecutive patients recruited over a 3.5-year period if they had a diagnosis of SCC or nerve root compression due to intraspinal metastases from a known solid malignant tumour. All 153 patients had SCC or nerve root compression confirmed by myelography (some received CT imaging). Patients were followed up from diagnosis of compression to death or for a minimum of 11 months. The primary cancer diagnoses were breast carcinoma in 56 patients (37%), prostatic carcinoma in 43 (28%), NSCLC in 18 (12%), SCLC in nine (6%), and other solid tumours in 27 (17%) patients. The distribution of compressions was cervical, seven (4%) cases; thoracic, 102 (67%) cases; and lumbosacral, 44 (29%) cases.

Results

The main results concerned predictors of overall survival and of ambulatory status after treatment for compression and these are outside the remit of this short report. The type of primary tumour was found to be a predictor of the time from diagnosis of cancer to the time of the first compression (p < 0.0005 in multivariate analysis), and also of the severity of the compression in terms of patients' ambulatory status at the time of diagnosis of compression. Of the tumour types examined, the shortest time to compression was found for lung cancer and the longest for breast cancer. Kaplan–Meier analyses of this outcome were not shown.

New compression at a different site was observed in 14 of 153 patients. In an analysis that lacked power because of small sample size, it was found that primary tumour type was not a predictor of recurrence. The median time to new recurrence after the first compression was 4.5 (range 1–25.4) months.

Author conclusions

There was a significant association (p = 0.016) between time interval from diagnosis of primary tumour until development of SCC and type of primary tumour. Pretreatment ambulatory function of SCC patients is a main determinant for post-treatment gait function. Survival time is short, especially in non-ambulatory patients, and can be improved only by restoration of gait function in non-ambulatory patients by immediate treatment.

Reviewer conclusions

Primary tumour type influences the time to SCC and patient walking status at time of confirmation of SCC. An inference that follows for studies with populations of mixed cancer type is that the length of time from primary diagnosis to study entry will influence the results of any analysis of prognostic factors predicting compression or vertebral fracture. There will therefore be considerable difficulties in interpretation of results from studies with a case-mix of patients with different cancer types and various delays between primary diagnosis and study entry.

Relevant aim

A stated aim of this study was to compare the risk of a recurrence of spinal cord or root compression among patients with different numbers of spinal metastases detected at the time of the diagnosis of the first compression.

Design and method

This was a prospective study of patients recruited over a 3.5-year period. The study included 107 consecutive patients with myelographically verified metastatic SCC or spinal root compression from a histologically verified solid tumour. The report states that all patients received radiotherapy after myelographic diagnosis, but also that ‘only those epidural lesions causing neurological signs or symptoms were irradiated’. The primary cancer diagnoses were reported as breast carcinoma in 42 patients, prostatic adenocarcinoma in 28 patients, lung cancer in 21 patients and other solid tumours in 16 patients. Multiple epidural lesions were observed in 37 (35%) patients; in one patient there were four separate lesions, in eight patients there were three lesions, and in 28 there were two separate lesions.

Results

Recurrence of compression was observed in 8 of 107 patients. There was no difference in risk of a second compression between patients with a single metastasis at the time of the confirmatory myelogram for first compression (five recurrence events among 70 patients at risk) and those with multiple metastases at the confirmatory myelogram (three recurrence events among 37 patients at risk). The overall survival was superior for those who experienced recurrence (n = 8, median 9.2 months) than for those with no recurrence (n = 99, median 3.5 months). Unsurprisingly, this indicates that a predictor for recurrence is prolonged survival and that identifying patients with recurrence tends to select those with longer survival.

Author conclusions

Only symptomatic epidural metastases should be irradiated, and all patients treated should be followed regularly and observed for a second SCC.

Reviewer conclusions

The number of recurrence events (n = 8) was too small to allow for meaningful investigation of prognostic factors predicting recurrence. Unsurprisingly, patients who survive longer are more at risk of recurrence.

Relevant aim

These authors aimed to assess indicators that might identify those patients with physician-suspected MSCC in whom MRI examination can be forgone.

Design and method

This was a prospective study of 280 consecutive patients, recruited over 2 years at a single centre, who underwent MRI for suspected SCC. Of 362 potentially eligible patients, 51 were excluded because they had MRI at other centres, and 31 were excluded because they did not undergo MRI for various reasons (e.g. unavailable scanner). The primary cancer diagnoses among the 280 included patients were various and reported as follows: breast 65, prostate 57, bronchus 72, haematological 23, urinary tract 21, gastrointestinal tract 13, unknown primary 12 and other 17 patients.

Patients received MRI, plain radiography and neurological assessment. The relative timing of the different imaging modalities was not reported.

Results

The presence of focal abnormality on radiographs together with neurological signs consistent with compression at that level was taken as positive diagnosis by radiography. On this basis, 104 out of 280 patients were judged positive; however, 13 had received previous radiotherapy at that site and were classified as MRI-mandatory because of previous therapy. The remaining 91 patients were classified as MRI non-mandatory. The remaining 176 out of 280 patients were negative by radiography plus neurological signs and were classified as MRI-mandatory.

Of the 280 MRI scans undertaken, 201 were positive for MSCC (186 extradural, five intradural but extramedullary and 10 intramedullary), and 79 were negative (of these, 19 showed no MRI abnormality and 60 showed various abnormalities). MSCC was observed at one level in 161 patients, at two levels in 36 patients and in three regions in four patients. The sites at which MSCCs were observed were cervical in 15 patients, thoracic in 160 patients and lumbar/sacral in 71 patients.

The diagnostic/prognostic performance of plain radiographs plus neurological examination for the diagnosis of MSCC was compared with MRI (the latter taken as gold standard), and specificity, sensitivity and positive and negative predictive values were calculated and reported. The results were presented in table 9 of the paper. Of the 91 patients classified as ‘non-mandatory MRI’ based on positive radiography plus neurological signs, 89 were positive for MSCC according to MRI, leaving two as MRI negative (i.e. false-positives). Of the 189 patients classified as ‘mandatory MRI’ based on negative radiography plus neurological signs (or by virtue of previous treatment), 112 were positive for MSCC by MRI. This should leave 77 true-negatives (189 − 112) by radiography; however, the table presents this number as 87, giving a total number of patients of 290 (10 more than included in the study). The reason for this discrepancy is unclear. The text states that for this analysis ‘thecal sac compression without SCC’ was viewed as a negative MRI, but it appears that these patients may have been double counted. The reported sensitivity, specificity and positive and negative predictive values were 44%, 98%, 98% and 44%, respectively. Values calculated on a total of 280 patients become 97% for specificity and 41% for negative predictive value.

Author conclusions

Although focal radiographic abnormalities with consistent neurological findings, when present, accurately predicted the presence and level of MSCC, whole-spine MRI is indicated in most patients with suspected MSCC because the additional information may alter the management plan. The primary tumour is not helpful in predicting which patients will have more than one site of compression, although this is uncommon in tumours of haematological origin.

Reviewer conclusions

The predictive/diagnostic performance of radiography plus neurological signs was poor (sensitivity only 44%), with more false-negatives than true-positives, but with a positive predictive value of 97%. However, predictive values are highly dependent on the prevalence of the condition in the population examined; here the prevalence was 69%, which tends to strongly favour a high positive predictive value, as illustrated in Figure 14.

FIGURE 14.

Strong dependence of positive predictive value (PPV) on prevalence. The curve shows the relationship between prevalence and positive predictive value when sensitivity and specificity are 44% and 97%, respectively, as in the study by Husband et al. 116 The straight line is when sensitivity and specificity are both set at 50%.

Relevant aim

The aim was to identify factors that might predict local recurrent disease (i.e. of spinal metastases).

Design and method

The main focus of this paper was to identify variables associated with prolonged survival or a favourable postoperative neurological status in patients who have received surgery for spinal metastases.

This was a prospective study of 101 patients (with a total of 106 spinal metastases) who received surgery for spinal metastases at a single hospital between 1977 and 1996. Patients were recruited from a total of 740 patients who had received spinal tumour treatment. The number of patients with spinal metastases not in receipt of surgery was unclear. The primary cancer diagnoses resulting in 106 spinal metastases were reported as breast 17, prostate 15, thyroid 9, kidney 12, unknown primary tumour 25, lung 17, colon 5, melanoma 2, urogenital tract 1, pleural mesothelioma 1, teratoma 1 and gallbladder 1. The time interval between cancer diagnosis and diagnosis of spinal metastases ranged from 2 days to 5 years (mean 4 months; SD 6 months). Spinal metastases were distributed as follows: cervical, 12 patients; thoracic, 62 patients; lumbar, 24 patients; and sacral, three patients. Various imaging methods were employed to establish the diagnosis of spinal metastasis. Health status was monitored according to the authors' published scoring system based on that of Karnofsky; scores ranged from 0 to 5.

Results

The 106 spinal metastases were subdivided according to primary tumour into ‘long survival prognosis’ (n = 53; breast, prostate, thyroid, kidney) and ‘short survival prognosis’ (n = 53; lung, colon, melanoma, urinogenital, mesothelioma, teratoma, gallbladder). The rationale for this subdivision was not elaborated. Kaplan–Meier analysis was used to investigate time to local recurrence and multiple logistic regression was used to identify influential risk variables for recurrence.

The absolute number of postsurgery local recurrences was not reported. According to Kaplan–Meier analysis, recurrence of spinal metastases leading to neurological deterioration (implying the presence of SCC) was observed in 57.9% of spinal metastases within 6 months of surgery, 69.3% within 1 year and 96% within 4 years (no risk table was provided; it is assumed that patients who died before recurrence were censored at time of death). Figure 3 of the study depicts the time to recurrence for ‘short survival prognosis’ and for ‘long survival prognosis’ patients; time to recurrence was much shorter for the latter. The reported relationship is represented in Figure 15. It is unclear if the prognostic subgroups were specified a priori. Similarly, time to recurrence was shorter for patients with better health status (< 3) than for those with scores ≥ 3. The relationship is represented in Figure 16.

FIGURE 15.

Time to local recurrence reported by Klekamp and Samii. 117 Data read from graph and fitted with Weibull distributions.

FIGURE 16.

Time to local recurrence reported by Klekamp and Samii. 117 Data read from graph and fitted with Weibull distributions.

Multiple logistic regression identified that long postoperative recurrence-free survival was associated with the following variables: favourable tumour histology (i.e. tumours in the long survival prognosis group category), tumours at cervical spine level, low number of affected vertebral bodies, good general health status, complete resection at surgery, and elective surgery (as distinct from emergency surgery; 70% of patients received emergency surgery). Adjuvant postoperative therapy, length of history and age did not show a significant influence on local metastatic recurrence rate.

Author conclusions

The authors' conclusions were largely based on a literature survey and concerned survival, treatment modalities and recommendations of treatment pathways as follows:

1. Patients in good health condition and living independently should undergo surgery for spinal metastasis if neurological symptoms are present. Postoperatively, adjuvant therapy should be initiated.

2. Patients with neurological symptoms but in poor condition, requiring hospitalisation for their cancerous disease independent of spinal metastasis, should not be operated on but should be offered radiotherapy and/or chemotherapy primarily.

3. Patients with spinal instability due to metastatic disease require stabilisation to achieve a satisfactory neurological outcome. However, a surgical procedure has to be tailored according to life expectancy and health status of the patient.

4. Patients without neurological symptoms or instability should undergo radiotherapy primarily.

5. Patients who deteriorate after or despite primary radiotherapy may be candidates for surgery, but more complications and higher mortality rates should be expected.

Reviewer conclusions

The patient population was very heterogeneous and the study spanned two decades, during which imaging and treatment modalities will have changed. Patients not judged suitable for surgery were excluded and therefore the study population represents physicians' judgements and may be particular to time and place. Unsurprisingly, type of primary tumour and patient's health status were identified as factors that influence the reappearance of spinal metastases after surgery. As recurrences were associated with neurological deficit it is probable that these factors will also be associated with SCC or vertebral collapse.

Relevant aim

The authors aimed to quantify the incidence of clinical signs and symptoms of malignant SCC, or of cauda equina compression, among patients with confirmed diagnosis of compression and to assess the utility of various imaging procedures for the diagnosis of malignant compression. The authors focused on the nature and onset of patient symptoms relative to the time of diagnosis, and on the reasons for delays in diagnosis.

Design and method

This was a prospective observational study of 319 patients with SCC or cauda equina compressions (with a total of 324 compressions) recruited at three centres between January 1998 and April 1999. Compression was mostly confirmed by MRI. The primary cancer diagnoses were reported as lung, prostate and breast, together accounting for 59% of all cases. Tumours were from the gastrointestinal tract in 10% of cases (32) while 10% were of haematological origin (myeloma, lymphoma, chronic lymphatic leukaemia) and in 23 cases (7%) the site of the primary tumour was never identified. Median age was 65 years and 203 out of 319 participants were male. The spinal level of compressions was reported to be cervical in 7% of cases, thoracic in 68%, lumbar in 21% and sacral in 4%. In 55 of 324 compressions, compression of more than one site was detected. In 72 out of 319 patients, the malignant compression was the presenting symptom; the remaining 247 patients were known to have cancer at the time of compression diagnosis. Diagnoses of malignant compression were mainly made on weekdays (average of 18.3% per day for all diagnoses) whereas weekend days only accounted for 4.3% per day, implying a few days' delay in diagnosis for a proportion of patients, presumably reflecting the lack of access to MRI outside the working week.

The proportions of patients with various clinical signs and symptoms of malignant compression, and in some instances the timing of their onset, were reported; however, no formal regression analysis of predictive factors or Kaplan–Meier analyses were undertaken. Various imaging methods were employed, including plain radiography, bone scintigraphy, CT and MRI. The accuracy of different modalities for diagnosis of malignant compression was compared.

Results

The reported frequencies of signs and symptoms are summarised in Table 24.

Pain was experienced by 94% of patients. Various categories of pain were common, and pain was found not to be the predictive factor of malignant cord compression; there was considerable discordance between the spinal level of pain and the structural level of compression. Eighteen per cent of patients were unable to walk by the time a diagnosis was made. There was no association between ability to walk and the patient's self-reported pain level (p = 0.99). Most clinical indicators were so common in this sample that they had little potential power as predictors.

The clinical level of sensory abnormality corresponded poorly with the level of cord compression identified on MRI scans. Considering the whole study population of 324 compressions, a sensory level was of value in identifying the level of compression in only 16%.

Factors contributing to delays in diagnosis of compression included slow general practitioner referral for patients not already known to have cancer, delay in referral after first appearance of signs or symptoms [median 66 days; interquartile range (IQR) 37–205 days], and delay between referral and definitive diagnosis (median 15 days; IQR 3–66 days). The rate of diagnosis of malignant cord compression increased through the week and was maximal on a Friday and low on weekend days.

Plain radiographs were obtained for 57% of compressions before diagnosis. Vertebral collapse (defined as ≥ 50% loss of vertebral height) was seen in 60 out of 187 (32%) plain films, yielding a sensitivity of only 32%. In 39 of these, the level of compression was confirmed on MRI. Hence, correctly predicted compression level was found in only 21% of radiographs. The most common request was for lumbar spine radiography, whereas the commonest site of compression was the thoracic spine.

Bone scintigraphy for back pain was performed in 139 patients. In 49, spinal hotspots suggestive of extensive bone destruction were identified, and in 26 of these the site corresponded to the level of compression as identified by MRI, yielding a true-positive rate of only 19% (26/139).

Author conclusions

Patients who developed spinal metastases were at risk of irreversible spinal cord damage. Weakness and sensory abnormalities were reported late and identified even later, despite patients having reported pain for a considerable time. Plain films and bone scans accurately predicted the level of compression in only 21% and 19% of cases, respectively. The only accurate investigation to establish the presence and site of a compressive lesion was MRI. Certain categories of patients are at risk of malignant cord compression, in particular, patients who are already known to have cancer when they first develop pain or who are > 50 years of age, and those with breast or prostate cancer with known bone metastases.

Reviewer conclusions

The paper looked at clinical symptoms, clinical signs and different technologies. The clinical symptoms and signs examined were found to be common in the selected population and therefore lacked discriminatory predictive power. MRI was judged to be the best available technology for detecting malignant compression and, relative to MRI, both plain radiography and bone scintigraphy were judged to perform badly.

Relevant aim

One aim was to identify and to assess the utility predictive models for MSCC, and similarly to compare imaging modalities for investigation of suspected MSCC (for previous more detailed discussion of concerns in relation to quality considerations for this systematic review, see Summary of systematic review evidence).

Design and method

The study was a systematic review that focused on the following seven questions concerning malignant SCC: (1) What are the clinical symptoms of MSCC? (2) What is the optimal approach for investigating suspected MSCC? (3) Is there a role for systemic corticosteroids in the management of MSCC, and if there is, what is the optimal dose? (4) What are the indications for surgery in the management of MSCC? (5) What are the indications for radiotherapy in the management of MSCC? (6) Is there an optimal dose prescription for radiotherapy? and (7) What are the treatment options for recurrent MSCC in an area previously irradiated?

Results

Fifty published studies were included and were reviewed by narrative description. Those publications that considered predictors of SCC or of vertebral collapse are discussed elsewhere in this short report.

Six studies were reviewed that addressed the relative utility of myelography and MRI for the investigation of suspected MSCC. These indicated that whole-spine MRI should be used for patients.

Author conclusions

Predictive risk models may help to define patients at higher risk of developing cord compression, but optimal screening strategy for a population and intervention have not been elucidated. Back pain was not predictive of MSCC. Treatment for patients with MSCC should consider presence of bony compression and spinal instability comorbidities, pretreatment ambulatory status, technical surgical factors, potential radiotherapy reactions, patient preferences and potential surgical complications.

Reviewer conclusions

Different factors such as inability to walk, increased deep tendon reflexes, compression fractures on radiographs of the spine, bone metastases present, bone metastases diagnosed > 1 year earlier and age < 60 years were found to be of some predictive value for MSCC. Back pain was found not to be predictive of MSCC.

Relevant aim

The aim was to identify independent clinical predictors of MRI-established SCC in cancer patients through the analysis of potential risk factors.

Design and method

This prospective study was a review of cancer patients with suspected SCC who were evaluated by MRI at two centres over the period from July 1998 to March 1999. Inclusion required consent by the physician ordering MRI. The patients included in the study had pathologically confirmed cancer diagnosis, no metastatic epidural cancer over the previous 12 months, age ≥ 18 years, and gave consent to a brief interview within 7 days of the scan. The patient interviews were conducted by one physician and focused on numerous factors experienced before MRI. Interviews were also conducted with the physicians ordering the scans; most patients were not blind to the results of the MRI. Of 167 eligible episodes of suspected SCC, a total of 136 episodes of suspected SCC among 134 patients were investigated by interview. The primary cancer diagnoses were reported as breast (n = 33; 24%), lung (n = 33; 24%), prostate (n = 21; 15%), non-Hodgkin's lymphoma (n = 8; 6%), multiple myeloma (n = 6; 4%) and others (n = 35, 26%). Median age was 61.5 (range 30.9–84.8) years.

Univariate analysis using Fisher's exact test was used to estimate the association of variables with a positive MRI test for SCC. Multivariate stepwise logistic regression was used to identify significant independent predictors of SCC.

Results

Clinically important metastatic epidural SCC was defined as any TSC with or without spinal cord displacement. MRI demonstrated 50 episodes of TSC reported at the following spinal levels: cervical 6%; thoracic 64%; lumbar 30%; sacral 6%. Vertebral metastases without TSC were seen in 46 episodes and no vertebral metastases in 40 episodes. In univariate analysis, back pain was not associated with TSC (92% of episodes were associated with back pain). Four independent variables predictive of TSC were identified by multivariate regression as follows: abnormal neurological examination (OR 3.0, 95% CI 1.6 to 10.4; p = 0.004); stage IV cancer at initial diagnosis (OR 2.8, 95% CI 1.4 to 7.7; p = 0.006); known vertebral metastases (OR 2.8, 95% CI 1.4 to 8.2; p = 0.008); and middle or upper back pain (OR 2.7, 95% CI 1.4 to 9.1; p = 0.010). These four predictors stratified patients experiencing episodes into subgroups with varying risks of TSC, ranging from 8% (no risk factors) to 81% (three or four risk factors), as summarised in Table 25. Only 19% of the episodes were associated with three or four risk factors relative to a prevalence of TSC of 36.7%.

Among the episodes not associated with abnormal neurological examination (n = 100), middle or upper back pain and stage IV cancer at initial diagnosis were found to be independent predictive variables. In this population these variables stratified patients experiencing episodes into subgroups with varying risks of TSC, ranging from 11% (no risk factors) to 69% (both risk factors), as summarised in Table 26. Only 13% of episodes in this population exhibited both risk factors.

Author conclusions

The results confirmed earlier retrospective studies indicating that evaluation of cancer patients with suspected SCC should be based on clinical information that includes cancer-related history, symptom data and presence of pertinent neurological signs. Predictors may help clinicians to assess risk in this patient population.

Reviewer conclusions

The selected population included several different types of cancers and so the four identified risk factors need to be tested in both similar and different mixed case populations to determine the generalisability of the findings. The primary tumour type and treatment with bisphosphonates or other interventions (except radiotherapy to the level of suspected SCC) which might influence the identity of predictive variables do not appear to have been included in the regression analyses.

Relevant aim

The aim of this study was to evaluate prospectively the probability of vertebral fracture in patients who have received single-dose image-guided intensity-modulated radiotherapy (IG-IMRT) for spinal metastases, and also to identify risk factors for such fracture.

Design and method

Image-guided-intensity-modulated radiotherapy is a recently developed therapeutic option for patients with spinal tumours. The authors noticed that a number of patients sustained vertebral fractures after IG-IMRT so they undertook a prospective study to monitor the incidence and risk factors for post-therapy fracture. The study included 71 lesions occurring in 62 patients given IG-IMRT for spinal tumours. The primary cancer types for the 71 lesions were reported as follows: renal cell, n = 14; melanoma, n = 9; prostate, n = 9; sarcoma, n = 7; colorectal, n = 6; cholangiocarcinoma, n = 5; thyroid, n = 5; non-small cell lung, n = 5; breast, n = 4; and other, n = 7. The spinal distributions of the treated lesions were reported as follows: cervical, n = 6; thoracic, n = 47; and lumbosacral, n = 18. The treated sites were classified on CT appearance as lytic (n = 46; 65%), sclerotic (n = 13; 18%) or mixed (n = 12; 17%). The sites were also classified according to the percentage of the vertebral body occupied by the lesion, as follows: 0–20% of the vertebral body occupied, n = 26 lesions (37%); 21–40% occupied, n = 18 (25%); 41–60% occupied, n = 10 (14%); 61–80% occupied, n = 7 (10%); and > 80% occupied, n = 10 lesions (14%).

After IG-IMRT patients were followed up using MRI at 2 months and then at 3-month to 4-month intervals. Multiple logistic regression and Cox's proportional hazard models were used to identify factors associated with fracture. Kaplan–Meier analysis was used to determine time to fracture. Fractures were classified as new or progressive; a progressive fracture represented a worsening, after IG-IMRT, of a lesion in which vertebral deformity or end-plate infraction existed at the time of IG-IMRT therapy. The following potential risk factors for fracture were examined: location of the lesion; size of the lesion (tumour occupancy in vertebral body); type of lesion (lytic, sclerotic or mixed); appearance of the lesion in CT; obesity; local kyphosis; bisphosphonate use; IG-IMRT radiation dose; presence of baseline fracture; and histology of fracture.

Results

Fracture progression was found in 27 vertebral bodies (39%). Multivariate logistic regression analysis showed that CT appearance (lytic or sclerotic/mixed), lesion location and amount of vertebral body occupied by tumour independently predicted fracture progression. Lesions located between T10 and the sacrum were 4.6 times more likely to fracture than were lesions above T10 (95% CI 1.1 to 19.7 times more likely). Lytic lesions were 6.8 times more likely to fracture than were sclerotic and mixed lesions (95% CI 1.4 to 33.3 times more likely). As the amount of vertebral body occupied by tumour increased, the odds of fracture increased.

Obesity, local kyphosis, bisphosphonate use, baseline presence of vertebral deformity and IG-IMRT radiation dose were not associated with increased risk of fracture. The presence of baseline fracture was not associated with new fracture development or progression. There was no clear correlation between histology and risk of fracture.

Median time to fracture taken from the Kaplan–Meier analysis was reported to be 25 months. Figure 17 illustrates the relationship and shows data read from the graph. A Weibull fit generated a median time to fracture of 25.02 months.

FIGURE 17.

Time to progressive fracture. Data read from graph (Rose 200980). Solid line represents a Weibull distribution fit to extracted data.

The median time to fracture for lytic lesions was reported to be 19 months, whereas the median time in sclerotic or mixed lesions was 32 months (p < 0.05). Figure 18 illustrates the time-to-event relationship and shows data read from the published graph. Weibull fits generated median times to fracture of 18.9 months for lytic lesions and 36.9 months for sclerotic/mixed lesions.

FIGURE 18.

Time to fracture for lytic and sclerotic or mixed lesions. Data read from published graph (Rose 200980). Numbers at risk assumed to be 46 and 25, respectively, and numbers of events assumed to be 23 and four, respectively. Solid lines represent fits using Weibull distributions; dashed line uses a Gompertz distribution and dotted line a log-log distribution.

By stratifying lesions according to location, median time to fracture changed significantly. The median time to fracture with lesions between T10 and the sacrum was 20 months, and for lesions located higher in the spine it was 35 months (p < 0.05). Stratification according to the amount of the vertebral body occupied by the lesion also resulted in significantly different fracture probability functions (p < 0.02).

In the multivariate proportional hazards regression model, only lytic appearance (HR 3.8, 95% CI 1.3 to 11.4) and lesions that occupied 41–60% of the vertebral body (HR 3.9, 95% CI 1.1 to 14.2) were associated with a statistically significant increase in the HR.

The Karnofsky performance status at final follow-up was 80%. The median change in Karnofsky performance status among patients with fracture progression was 10%, and among patients without fracture progression was 0% (p < 0.03).

Author conclusions

The study identifies a high risk of vertebral fracture after single-fraction IG-IMRT to spinal metastases. Lytic disease involving > 40% of the vertebral body and location at or below T10 confers a high risk of fracture, the presence of which yields significantly poorer clinical outcomes.

Reviewer conclusions

The study explores fracture risk after single-fraction IG-IMRT treatment. Risk of progressive fracture after IG-IMRT was appreciable; poor prognosis for fracture appeared to be associated with lytic lesions, those occupying > 40% of the vertebral body and greater load on the deformed vertebra (as indicated by greater risk for lesions below T10). This was a small study with potentially very important findings but because of quality considerations its validity is difficult to evaluate.

Relevant aim

The study aimed to assess the predictive utility of biomechanically derived models to accurately predict the risk of vertebral burst fracture in the metastatic spine, and to generate simple methods to obtain the required data needed to make such risk assessment of burst fracture.

Design and method

This was a retrospective study of all cancer patients seen at a single centre between September 1998 and November 2001 who, on the basis of available CT imaging, were considered to have osteolytic metastases of the thoracic or lumbar spine. Vertebrae were classified as not fractured or as bearing burst or wedge fractures.

Of 560 potentially eligible patients with spinal metastases, 117 had suitable CT imaging and, of these, 72 (34 male and 38 female) harboured osteolytic spinal metastases (48 thoracic and 44 lumbar) and were included. Of the 92 metastatic vertebrae, 21 (23%) harboured fractures (17 burst fractures and four compression fractures), and 71 (77%) were not fractured. The primary cancer diagnoses were reported as follows: breast, n = 23; lung, n = 7; colon, n = 3; prostate, n = 5; lymphoma, n = 6; multiple myeloma, n = 5; renal, n = 4; other, n = 10; and unknown, n = 9.

The following estimates were made for each vertebra: vertebral body volume; minimal cross-sectional area; tumour volume (as a percentage of the vertebral body volume); apparent bone mineral density; tumour volume in the pedicle (dichotomised as intact or involved) and intervertebral disc quality (dichotomised as healthy or degenerated); pressure load based on patient weight, activity level and apparent cross-sectional area of the vertebra; and estimated proportion of body weight above the vertebral level. Loading rate, dichotomised as high or normal, was also recorded. For fractured vertebrae the minimal sectional area was estimated from that of adjacent intact vertebrae. The data estimates were used in biomechanical models, developed in a previous study, so as to determine the risk of burst fracture for each vertebra. The predicted outcome could then be compared with the known presence or absence of a burst fracture.

Results

The most accurate predictor of burst fracture was a model of vertebral bulge using only the spinal load-bearing capacity (constant pressure load). At an appropriate threshold (5.04, with a margin of 0.37) this had sensitivity and specificity of 1 for distinguishing burst-fractured vertebrae from unfractured or wedge-fractured vertebrae, and in logistic regression a Hosmer–Lemeshow test value of 1. Burst fracture prediction using vertebral axial displacement and tumour size were also good predictors. None of the models performed well at distinguishing unfractured from fractured (burst or wedge) vertebrae.

Author conclusions

Fracture prediction was optimised using the vertebral bulge model considering only load-bearing capacity with a specificity, sensitivity and CI of 1 to yield a clear threshold for burst fracture risk. Fracture prediction in the other two models, vertebral axial displacement considering only load-bearing capacity and tumour size, was also strong, with receiver-operator curve values of 0.992 and 0.988, respectively. The predictive power of these models can provide useful clinical information for prophylactic decision-making.

Reviewer conclusions

As indicated by the authors, the operator inputs required to undertake the modelling described are considerable and the methods used required relatively sophisticated digital scanning equipment, which may not be widely available. The development of automated systems may be required for the necessary data collection to become routine. Although prediction of burst fractures was impressive, the number of samples included was small and the validity of the results needs testing in a larger sample and in different populations.

Relevant aim

The aim of this study was to identify risk factors for metastatic vertebral fracture and epidural impingement.

Design and method

This was a retrospective study of metastatic cancer patients with spinal metastases. Patients were excluded if the primary tumour was a myeloma, lymphoma or other tumour of haematopoietic origin, if MRI was carried out within 30 days of a surgical intervention or if MRI demonstrated a metallic implant.

A random sampling method was reported and was used to select two samples from a population of MRI-evaluated patients with spinal metastases seen at one university hospital between October 1992 and June 1998. The first sample was used to estimate the incidence of vertebral fracture and risk factors for fracture; the second sample was selected from patients who presented with vertebral fracture and was used to investigate progression from normal shape, patterns of fracture and progression to epidural impingement.

The first sample comprised 53 patients (mean age 58 years, SD 26 years; 26/53 male) with images by MRI of 756 vertebrae. The primary tumours were reported as follows: breast, n = 14 patients (26.4%); lung, n = 13 (24.5%); prostate, n = 9 (17%); renal, n = 7 (13.2%); undifferentiated, n = 3 (5.7%); and others, n = 7 (13.2%). Metastatic lesions were observed in 253 out of 756 vertebrae (33.4%). In 114 of these, an isolated zone of lysis or of osteoblastic new bone could be identified; these lesions were classified as circumscribed. Of circumscribed lesions, 104 (91.2%) were confined to the vertebral body (excluding arch and pedicles). The metastatic lesions were observed most commonly among lumbar and posterior thoracic vertebrae.

The second sample comprised 67 patients presenting with vertebral fractures (113 fractured vertebrae). Twenty-two fractures were found to have no metastatic infiltration and were not analysed further, leaving a final sample of 91 fractured vertebrae.

Results

The risk of vertebral fracture among infiltrated vertebrae was greatest for upper lumbar (L1–L3) vertebrae relative to other vertebrae (RR 1.95, 95% CI 1.12 to 3.38; p = 0.017), and for undifferentiated tumours relative to other tumours (RR 7.36, 95% CI 2.69 to 20.12; p = 0.001). Prostate metastases were associated with the smallest risk of fractures (RR 0.21, 95% CI 0.082 to 0.535; p = 0.001).

MRI follow-up identified 23 normally shaped vertebrae that progressed to fracture. According to a Cox's proportional hazards model, greater fracture risk was noted in vertebrae with > 80% vertebral body infiltration relative to less infiltration (HR 4.60, 95% CI 1.66 to 12.71).

The number of spinal levels affected by metastasis was weakly correlated with the number of fractured vertebrae in an individual patient (r = 0.325). There was no significant correlation between metastatic involvement of one or both pedicles with fractures (p = 0.43).

Fractures were classified by the authors as (1) symmetrical compression wedge fracture (with ‘delta fragments’); (2) symmetrical compression fracture with no ‘delta fragments’; (3) lateral compression fracture; or (4) anterior compression fracture. Those classified as wedge fractures had a greater tendency to progress to migration into the epidural space.

Author conclusions

Fracture risk was greatest for upper lumbar and undifferentiated tumours. Fracture risk was substantially increased in vertebrae with > 80% body infiltration, and symmetrical fractures with fragments were associated with the greatest risk of epidural impingement.

Reviewer conclusions

The reported results are perhaps unsurprising in that those vertebrae tending to bear greater load and sustaining greater metastatic infiltration are more likely to fracture. Similarly, fractures generating bony fragments might be expected to cause more serious epidural penetration. The finding that risk of fracture appears to vary with primary tumour diagnosis is of interest for this report in that it implies that results with mixed cancer type populations might be viewed as largely reflecting the proportional contribution of the different cancer types.

Relevant aim

The aim was to examine potential clinical neurological and oncological risk factors for CT-established SCC in metastatic cancer patients with suspected SCC.

Design and method

This was a retrospective study of medical and CT scan records accumulated between 1 February 1985 and 30 September 1988 at a single centre. Patients were included if a CT scan had been conducted for clinically suspected SCC (where SCC = SCC or cauda equina syndrome); this was termed the index scan. Patients were excluded if CT was not carried out for suspected SCC or if they had a previous diagnosis of SCC.

Of 405 index scans identified from records, 342 (in 258 patients) were included for analysis. The reasons for exclusion of 63 scans were reported. Mean age at study entry was 56.5 years. Primary tumour diagnosis was reported as breast in 42% of patients; NSCLC in 14%; prostate in 9%; sarcoma in 5%; and other in 30%.

The time period, study centre, number of index scans identified and the number excluded (n = 63) correspond to the Lu et al. 1998 study120 (described earlier); however, Lu et al. investigated only breast cancer patients (number reported as 93). Talcott et al. report on 258 patients, of whom 42% had diagnosis of breast cancer (n = 108); it is likely that most of the breast cancer patients in this study are identical to those reported by Lu et al. 120

The spinal level of the suspected episodes was reported as L3 or L4 in 43% of 342 index scans and T13 in 30% of 342 index scans. Uncertain scans (< 5%) were followed up by myelography or MRI. Most patients received imaging before the index CT, mostly to document metastases to bone, especially spine. Plain film radiographs immediately preceded 250 of the 342 index scans; vertebral lesions (lytic 29%, blastic 16%, mixed 20%) were seen in 68% of the plain films and compression fractures were seen in 30%.

Results

Twenty-two variables were examined in univariate or multivariate logistic regression for association with SCC. Several definitions of SCC were employed: TSC; spinal cord or cauda equina displacement (SCD); TSC + SCD; EM; SCD + TCD + EM.

A positive diagnosis at index scan depended on the formal definition of SCC used. For TSC, 29 out of 342 scans were positive; for SCD, 43 out of 342; for EM only, 52 out of 342; for TSC + SCD, 72 out of 342; for TSC + SCD + EM, 124 out of 342; and for TSC + SCD at index or within the 90-day follow-up, 80 out of 342. If local radiation at the site of suspected SCC within 90 days of a negative index CT, as an indication of SCC, then 169 out of 342 (49%) index scan episodes were positive. The reported associations of variables with SCC in univariate logistic regression is summarised in Table 27.

In multivariate logistic regression six variables were significantly associated with TSC at index or during 90 days of follow-up. These are summarised in Table 28.

The authors counted the number of these six risk factors present in each patient and related this to the occurrence of TSC at index scan or within 90 days. The results are summarised in Table 29 together with the post-test probability of having TSC. According to these results, the 23% pre-test probability of TSC (within 90 days of the index scan) is raised to 87% for a patient who exhibits five risk factors.

Author conclusions

The clinical history of patients' cancer contributes independently to risk assessment. The prevalence of SCC depends on definition used and whether short-term clinical follow-up is included.

Reviewer conclusions

The authors' main conclusion is justified. The 23% prevalence of TSC among the CT index scans may not be surprising because patients were selected for suspected SCC. Several of the risk factors identified were also probably unsurprising as they included vertebral fracture on most recent radiograph (250 of the 342 index scans were immediately preceded by plain radiograph and fracture is known to be highly associated with SCC); previously diagnosed bone metastases (these may have progressed for a long time before the index scan, and although breast cancer and other patients may have osteoporosis, which might independently lead to bone fractures, it is unlikely that SCC will occur without bone metastasis in this selected population); complaint of inability to walk (a well-known symptom of SCC); increased deep tendon reflex; and age < 60 years. Spinal imaging has advanced since this study was conducted (1985–8) so that some of the findings might be interpreted differently in the current context.

Relevant aim

In an attempt to accurately diagnose impending collapse of vertebrae with metastatic invasion, the authors determined the risk factors for vertebral collapse, estimated the predicted probability of collapse under various states of metastatic vertebral involvement and established criteria for impending collapse.

Design and method

Fifty-three patients, with (n = 40) or without vertebral collapse, harbouring 100 thoracic or lumbar metastatic tumours were studied. The patients' average age was 59.7 years (SD 8.8, range 43–80 years). The sampling frame was not reported but it is implied that patients visited a single centre. The vertebrae were selected if they satisfied the following conditions: they contained purely or predominantly osteolytic metastatic lesions; there were no end-plate fractures in adjacent vertebrae; tomography (sagittal and coronal plane) and CT had been performed within 1 week of an initial plain radiographic examination, and images were judged to be qualified for detailed analysis; and radiographic examinations were performed before biopsy, radiation therapy or surgical treatment (e.g. laminectomy).

A variety of primary cancers were reported to be associated with the 100 vertebrae, as follows: renal cell carcinoma, n = 17; breast, n = 15; prostate, n = 15; hepatocellular, n = 13; lung, n = 8; oesophageal, n = 4; thyroid, n = 3; gastric, n = 3; colon, n = 2; melanoma, n = 2; fibrous sarcoma, n = 1; rhabdosarcoma, n = 1; leiomyosarcoma, n = 1; lymphoma, n = 1; ureter, n = 1; adrenal, n = 1; and unknown, n = 12. The distribution of affected vertebrae was as shown in Figure 19 and was reported as equally split between T4 to T10 (n = 50) and T11 to L4 (n = 50).

FIGURE 19.

Distribution of metastatically affected vertebrae in the study by Taneichi et al. 89

The potential risk factors for vertebral collapse examined reflected size and disposition of the metastatic lesion in the affected vertebra. Four factors were estimated. Factor [1] was the percentage tumour occupancy (% TO) of the vertebral body. To estimate %TO the CT images were examined using computer software so as to gain measures of A, the most extensive cross-sectional area of the tumour, and B, the cross-sectional area of the adjacent unaffected vertebra measured in the same plain as A. The %TO then = 100 × A/B; if A in the affected vertebra could not be reasonably measured then the intact part of the vertebral body C was measured, and a value for A was obtained indirectly as B–C. Factors [2], [3] and [4] were dichotomised according to CT image estimates of the presence of destruction of, respectively, the pedicle, the posterior elements (not including the pedicle); and the costovertebral joint (T vertebrae only).

Vertebral collapse was defined as (1) a vertebra with fractures of the end-plate adjacent to the osteolytic lesion, and (2) a vertebra with reduction of vertebral body height because of pathological fractures of the anterior and/or lateral cortex of the vertebral body. Vertebral body height was considered to be reduced when the height of the affected vertebral body was < 90% of the estimated original height. This was calculated from an average of the corresponding measurements at adjacent unaffected levels above and below the metastatic vertebra.

The authors developed a multivariate logistic regression model to determine the associations between the occurrence of vertebral collapse and the four risk factors. The predicted probability of vertebral body collapse in various states of metastatic vertebral involvement was estimated using the developed model. A set of criteria for ‘impending vertebral body collapse’ was suggested.

Results

There were no differences between T and L groups in the frequency of risk factors in collapsed and intact vertebrae, nor in the %TO of collapsed vertebrae (Table 30).

The results of multivariate logistic regression for the T and L group vertebrae are summarised in Tables 31 and 32.

For group T vertebrae, the strongest association was between costovertebral joint destruction and vertebral collapse (OR 10.17; p = 0.021). The tumour size (%TO) was also associated with the risk of vertebral collapse (p = 0.032) with an OR of 2.44 for every 10% increment in %TO. Destruction of the pedicle and other posterior elements was not associated with collapse (OR 1.73; p = 0.703, and OR 1.17; p = 0.886, respectively).

For group L, the two most important risk factors for vertebral body collapse were %TO (OR of every 10% increment in %TO 4.35; p = 0.002) and pedicle destruction (OR 297.08; p = 0.009). Destruction of the posterior elements was inversely correlated with the risk of collapse (OR 0.03; p = 0.027).

Based on these results the authors developed the following equations to describe the probability of vertebral collapse in group T and group L:

where [1], [2], [3] and [4] refer to risk factors listed in Table 32 and take values [1], 0–100 and [2], [3] and [4], 0 or 1.

where [1], [2] and [3] refer to risk factors.

Examples of the reported predicted probabilities of fracture for the T group, calculated according to the equation above, are shown in Table 33.

Similar reported predicted probabilities of fracture for the L group are shown in Table 34.

On the basis of these predictors, the criteria of impending collapse in group T were defined by the authors as (1) 50–60% (%TO) involvement of the vertebral body with no destruction of the other structures; and (2) 25–30% (%TO) involvement of the vertebral body with costovertebral joint destruction. In group L, the criteria were defined as (1) 35–40% (%TO) involvement of the vertebral body with no destruction of the other structures; and (2) 20–25% (%TO) involvement of the vertebral body with destruction of the posterior elements including the pedicle.

Author conclusions

With respect to the timing and occurrence of vertebral collapse, there is a distinct discrepancy between the thoracic and thoracolumbar or lumbar spine. When a prophylactic treatment is required, the optimum timing and method of treatment should be selected according to the level and extent of the metastatic vertebral involvement.

Reviewer conclusions

This study is more complete than many in that it develops empirical equations for prediction of fracture. The study selected only intraspinal tumour-related factors as risk factors for collapse; extraspinal factors such as age and sex were not considered. Any effect exerted from different primary types was not explored. Intraspinal factors such as costovertebral joint destruction and tumour size in the thoracic region were found to be significant risk factors. Factors such as tumour size and pedicle destruction were found to be significant risk factors in the thoracolumbar and lumbar spine. The equations developed need testing prospectively in different populations with spinal metastases.

Relevant aim

A stated objective of this study was ‘To use an evidence-based medicine process using the best available literature and expert opinion consensus to develop a comprehensive classification system to diagnose neoplastic spinal instability’. The authors comment that spinal stability may be defined as the ability of the spine to maintain its degree of motion while simultaneously preventing pain, neurological deficit and abnormal angulation.

Design and method

The authors (34 members of a Spine Oncology Study Group) used evidence provided by two systematic reviews undertaken by members of the group, and attempted to integrate this with expert opinion through a modified Delphi technique to generate through consensus of best evidence and expert opinion a classification system to define neoplastic spinal instability. The outcome was a spinal instability scoring system (Spine Instability Neoplastic Score). The authors believe that the Spine Instability Neoplastic Score, together with a patient's overall prognosis and physical condition, should be taken into account when considering interventions (e.g. surgery) that might be appropriate.

One systematic review was referenced as ‘in press’: (1) Fehlings MD, Furlan J, Bilsky M, et al. Defining oncologic instability of the cervical spinecan the available evidence guide clinical practice? Spine (in press). This study was not retrieved in our searches. A study by Weber et al. 52 was excluded from this short report because the outcome measures did not meet the inclusion criteria (see Appendix 5). Most of the studies included in the Weber systematic review were conducted with animal models, cadaver vertebrae or computer modelling and were not concerned with testing prognostic variables in humans with spinal metastases.

Results

The authors developed a Spine Instability Neoplastic Score for a series of variables that may be present in a particular patient. The variables were classified under six categories and are summarised in Table 35.

Author conclusions

The Spine Instability Neoplastic Score was found to be a comprehensive classification system with content validity that could potentially guide clinicians in identifying when patients with neoplastic disease of the spine may benefit from surgical consultation. The Spine Instability Neoplastic Score might aid surgeons in assessing the key components of spinal instability due to neoplasia and may become a prognostic tool for surgical decision-making when put in context with other key elements such as neurological symptoms, extent of disease, prognosis, patient health factors, oncological subtype and radiosensitivity of the tumour.

Reviewer conclusions

It is difficult to discern how the findings of the systematic reviews fed into the study findings. In essence, this study is an expert opinion piece, and as such its quality is difficult to gauge using conventional assessment procedures. As with other studies included in this short report, the potentially prognostic variables identified require testing quantitatively in appropriate relevant populations; the clear difference between this study and others is that the prognostic variables deemed important have been identified on the basis of evidence synthesis by expert opinion.