Percutaneous vertebroplasty and percutaneous balloon kyphoplasty for the treatment of osteoporotic vertebral fractures: a systematic review and cost-effectiveness analysis

Quality of life

In people with osteoporotic fracture, quality of life can deteriorate quickly, even when physical function is not drastically affected, if changes in physical appearance, fear of fracture, and impediments to social function cause loss of self-esteem. 10 While most of the relevant research has been performed in post-menopausal women, men with VCFs and primary or secondary osteoporosis attending a UK hospital bone clinic scored much more highly in all six domains of the Nottingham Health Profile than age-matched or elderly male control subjects; the difference was particularly marked for energy, pain and physical mobility. The physical mobility scores indicated greater disability in men with secondary osteoporosis than in those with primary osteoporosis (p < 0.05). 47

Fears, anxiety and depression

Symptomatic VCFs are associated with fears, anxiety and depression, which may relate to fear of future fractures, fear of loss of independence, and a feeling of hopelessness resulting from being told to avoid activities such as bending, twisting and lifting heavy items, without being given advice on how to compensate. 10 Although post-menopausal women with a single VCF retain a good quality of life, once they have more than one fracture their quality of life is adversely affected by high levels of anxiety largely caused by fear of future fractures. 38 Such anxiety often leads to inactivity, which in turn can exacerbate BMD loss and declines in physical fitness, thus increasing the risk of falling. 10,38

In a small case–control study, Lyles et al. 39 found that psychiatric symptoms, as measured by the Hopkins Symptom Checklist Revised (SCL-90-R), were significantly worse in women with VCFs (p = 0.043) than in matched control subjects; however, there was no significant difference in depression as measured by the Beck Depression Inventory (p = 0.129). Cook et al. 43 found that emotional problems were common in post-menopausal women with chronic back pain due to VCF: 82% reported fear of falling, while 66% reported frustration and 53% reported anger. Unfortunately, this study did not include a control group of similar women without chronic back pain due to VCF.

Self-esteem

Vertebral compression fractures may lead to height loss, spinal deformity and abdominal protrusion, which adversely affect self-image and self-confidence, and to functional limitations which may lead to a loss of self-esteem by limiting independence and the ability to participate in social activities. 38 Even relatively mild chronic pain may cause discomfort which discourages participation in social activities that involve sitting or standing for extended periods. Moreover, spinal curvature and height loss may make it difficult or impossible to sit or stand erect, causing problems with conversation and other activities. 10 Cook et al. 43 found that over 50% of post-menopausal women with a diagnosis of chronic back pain due to osteoporotic VCF reported problems with leisure/social activities. However, in a small case–control study, Lyles et al. 39 found that women with VCFs and matched control subjects did not differ significantly in self-esteem as measured by the Rosenberg Self-Esteem Scale (p = 0.731).

Social support and social roles

The pain and physical impairment caused by VCFs can undermine the reciprocity involved in interpersonal relationships by reducing the ability to provide help and support to family and friends, while potentially increasing the need for assistance with activities of daily living and other personal care. If people are obliged to give up work, domestic, recreational or sexual activities because of the limitations on their physical and functional abilities, they may also be deprived of their social roles. 38 The impact may be severe even if the activities in question do not seem to others to be demanding; the inability to stand or sit for extended periods may limit involvement in social events, leading to an inability to fulfil the social roles that form an important source of self-esteem, and thus to a severe reduction in quality of life. 10

Sources searched

The following electronic databases were searched from inception to 22 November 2011: MEDLINE (Ovid); MEDLINE In Process & Other Non-Indexed Citations; Cumulative Index to Nursing and Allied Health Literature (CINAHL); EMBASE; EconLit; The Cochrane Library including the Cochrane Database of Systematic Reviews, Cochrane Controlled Trials Register (CCTR), Database of Abstracts of Reviews of Effects (DARE), NHS Economic Evaluation Database (NHS EED) and Health Technology Assessment (HTA) databases; and Science Citation Index (SCI).

Search terms

The search strategy was developed in collaboration with the information specialist. Search terms included ‘vertebroplasty’, ‘kyphoplasty’, and a broad variety of related clinical terms (e.g. ‘bone void fill*’, ‘vertebral* and augmentation*) in order to obtain a wide scope. No bibliographic filters were used. Vocabulary around vertebroplasty/kyphoplasty is limited and, therefore, few synonyms were available. The searches were simple with an emphasis on sensitivity, utilising both keywords and MeSH/thesaurus terms where available. The MEDLINE search strategy is provided in Appendix 2 . Search strategies for the other databases are available on request.

Search restrictions

Searches were not restricted by language, publication date or publication type (with exception of removing letters, news, editorials, etc.). Furthermore, they were not restricted by study design, because studies that did not meet the inclusion criteria for the review of clinical effectiveness could have provided relevant information relating to adverse events or been important in identifying further relevant papers and current research.

Population

The population comprised people of any age and either sex with painful osteoporotic VCFs. Studies which also included participants with non-osteoporotic vertebral fractures of other aetiologies (e.g. fractures associated with trauma, myeloma or metastatic cancer) were included if data relating to participants with osteoporotic fractures could be extracted separately, or if the proportion of participants with non-osteoporotic fractures was extremely small.

Intervention(s)

Percutaneous vertebroplasty; BKP with or without vertebral body stenting.

Comparator(s)

The interventions themselves, non-invasive management, OPLA, or no treatment.

Outcomes

The primary outcomes of interest for this appraisal were:

• HRQoL

• back-specific functional status/mobility

• pain/analgesic use

• vertebral body height and angular deformity

• progression of treated fracture

• incidence of new vertebral fractures.

Secondary outcomes were:

• all-cause mortality

• symptomatic and asymptomatic leakage of cement (e.g. into adjacent intervertebral discs)

• peri-procedural balloon rupture

• postoperative complications (including infection)

• other adverse events

• resource utilisation

• cost utility.

Only studies which reported data relating to at least one of the primary outcomes listed above in relation to the population of interest were eligible for inclusion in the review of clinical effectiveness. This criterion was relaxed for consideration of adverse events, to allow the inclusion of studies which reported data relating to any of the secondary outcomes in the population of interest. However, adverse event data were included only if they related to patients with osteoporotic VCFs because of the possibility that patients with fractures of different aetiology (e.g. malignancy) might be susceptible to more, or different, adverse events.

To facilitate comparison, outcomes measured at or before 3 weeks are grouped together as short-term outcomes, those measured between 1 month and 6 months as medium-term outcomes, and those measured at 12 months or later as long-term outcomes.

Study design

According to the accepted hierarchy of evidence, the review of clinical effectiveness was limited to RCTs, as they provide the most authoritative form of evidence. It was planned that non-randomised studies would be considered if insufficient data were available from RCTs, but this was not necessary.

In reviews of interventions for which beneficial effects are uncertain or contentious, with some possibility of harm, an accompanying review of adverse events can be of substantial importance when deciding whether or not to use the intervention. 120 It is widely recognised that RCTs do not form a good source of evidence for adverse events: they are generally not powered to reliably detect rare adverse events, nor is their follow-up period long enough to permit the detection of adverse events widely separated in time from the original intervention. 121 In addition, their populations are often not wholly typical of the target population; they may be younger and have fewer comorbidities than the general population of patients with the condition of interest. 122 Moreover, RCTs do not always measure all potential side effects. 123 Hence, it was decided to review the literature on adverse events in PVP and BKP to provide additional support for clinical decision-making. Adverse events were addressed using two broad research questions, namely, ‘what adverse events are associated with PVP or BKP in the treatment of osteoporotic VCFs?’ and ‘what is the approximate incidence of adverse events associated with PVP or BKP in the treatment of osteoporotic VCFs?’ Although this broad approach risked an overload of heterogeneous data which could not be easily pooled, it had a twofold advantage: first, it could identify new or previously unrecognised complications, and second, it could provide a more comprehensive overview of potential complications. 120

Two types of evidence were included in the review of safety:

• Large observational studies (≥ 200 patients), which would allow exploration of the range and incidence of adverse events associated with PVP and BKP. The decision to include only large observational studies was based on a desire to exclude small case series which might display particularly high adverse event rates associated with limited experience of the relevant techniques on the part of the clinician or institution. The decision to set the threshold for inclusion at 200 patients was taken a priori.

• Individual case reports were used to supplement the RCTs and large observational studies to provide as full a picture as possible of the range of adverse events associated with PVP and BKP. They were therefore used as a source of evidence relating only to adverse events which were not reported in the RCTs or large observational studies. By their nature, individual case reports cannot provide any indication of the incidence of such adverse events.

As with the review of clinical effectiveness, studies which included participants with non-osteoporotic vertebral fractures of other aetiologies (e.g. fractures associated with trauma, myeloma or metastatic cancer) as well as those with osteoporotic VCFs were excluded unless data relating to participants with osteoporotic fractures could be extracted separately. This was because there is some evidence that the type and incidence of adverse events may differ in vertebral fractures of non-osteoporotic origin (e.g. metastatic or traumatic). 124,125

Study selection

Retrieved studies were selected for inclusion through a two-stage process according to the above inclusion/exclusion criteria. The references identified by the literature searches were assessed for relevance first by title/abstract and then by full text, excluding at each step studies which did not satisfy those criteria; abstract-only publications were retained for full-text review. One reviewer examined titles and abstracts for inclusion, and screening was checked by a second reviewer on 10% of citations. For studies of clinical effectiveness, Cohen’s kappa coefficient (range 0–1) calculated to measure inter-rater reliability was excellent, at 1.0, indicating no discrepancies.

Data extraction strategy

Data were extracted independently by two reviewers using a standardised data extraction form (see Appendix 3 ); discrepancies were resolved by discussion and did not require input from a third reviewer. Where multiple publications relating to the same study were identified, data were extracted and reported as a single study.

Data obtained from the submissions made by the manufacturers have been appraised and commented on where deemed relevant.

Critical appraisal strategy

The methodological quality of each study that met the inclusion criteria for the review of clinical effectiveness was assessed independently by two reviewers, and any discrepancies were resolved by discussion. Where a study was reported in more than one publication, its quality was assessed on the basis of the combined data from all relevant publications.

It was stated in the protocol that quality would be assessed according to criteria based on those proposed by Ploeg et al. 88 for the assessment of studies of PVP (see Appendix 4 ). These criteria were initially adopted because they could be applied to both randomised and non-randomised studies. However, in the event, because sufficient RCTs were identified, it was not necessary to include non-randomised studies, and it was found that the criteria proposed by Ploeg et al. did not discriminate sufficiently between the included RCTs. A new set of criteria was therefore developed; this was based on the criteria proposed by Centre for Reviews and Dissemination and The Cochrane Collaboration for assessing the risk of bias in randomised trials, but also incorporated criteria proposed by Ploeg et al. 88 and Furlan et al. 126 which had particular relevance to the interventions under review. These criteria relate to internal validity, and also to external validity, and precision (for details, see Appendix 5 ). The criterion relating to the blinding of care providers was excluded as such blinding was not possible given the nature of the interventions under review.

The revised quality assessment tool included some questions that led to subsidiary questions to which the answer could be ‘not applicable’. These subsidiary questions have not been included in the risk of bias tables presented later in this chapter (see Study characteristics).

Methods of data synthesis

Owing to the potential impact of baseline imbalances in the degree of pain and disability reported by patients with osteoporotic VCFs, it is crucial that outcomes which are reported as continuous data (e.g. pain, disability and HRQoL) are assessed in terms of the difference between the mean changes from baseline in the intervention and control groups, and not in terms of the differences between mean scores at any given point in time. Where the original study investigators presented relevant measures of effect in terms of mean changes from baseline, these have been included in the data tables not least because in some cases they also adjusted for stratification variables (e.g. treatment centre). Where adjusted data were not reported, mean between-group differences in change from baseline for continuous outcomes were calculated adjusting for the variance of the within treatment change from baseline, where this was made possible by the data. This method generated CIs but not p-values. For dichotomous outcomes, relative risks, with CIs and p-values, were calculated using The Cochrane Collaboration’s Review Manager software (version 5.1, The Nordic Cochrane Centre, Copenhagen, Denmark) if such data were not reported by the study investigators.

Studies which met the review’s entry criteria were eligible for inclusion in meta-analyses to estimate summary measures of effect if such meta-analysis was appropriate (i.e. if the study populations, intervention and outcomes were comparable). Meta-analysis was carried out using random effects models, using Review Manager. Heterogeneity in the meta-analyses was explored through consideration of the study populations, methods and interventions by visualisation of results and, in statistical terms, by the chi-squared test for homogeneity and the I ²-statistic. However, such meta-analysis was limited to dichotomous outcomes.

The review team did not undertake meta-analyses of data relating to continuous or quasi-continuous outcomes. Such meta-analysis was considered inappropriate because of the existence of a published meta-analysis by Staples et al. 110 of data from the only two double-blind studies of vertebral augmentation (Buchbinder et al. 102 and INVEST103). As this meta-analysis used IPD, it was of a higher quality than could be achieved using published data. There was considered to be too much heterogeneity to justify combining data from all the studies of PVP in a meta-analysis together with published data from the Buchbinder and INVEST studies.

Health-related quality of life

Generic measures of health status are particularly important in populations with comorbidities, such as the elderly, because disabilities from these comorbidities may influence the patient’s response to treatment. Also, because such measures include mental and social health, they give a more complete picture of the patient’s health than do back-specific instruments. 149 Given the perspective of the research, the measure of most relevance was the EQ-5D. This is detailed first, followed by other HRQoL instruments in alphabetical order.

The EQ-5D has five dimensions (mobility, self-care, usual activities, pain/discomfort and anxiety/depression) each assessed by a single question on a three-point ordinal scale (no problems, some problems, extreme problems). These are responses are combined and presented as a quasi-continuous outcome on a scale of –0.59 to 1.00, where 0 represents death and 1.00 indicates ‘full health’; negative scores represent health states valued as worse than death. The estimated MCID for people with back pain is 0.08. 150

The Assessment of Quality of Life (AQoL) measure was designed for use in the evaluation of health-care interventions, and is sensitive to changes in the frail elderly. It is a multiattribute quality of life utility instrument, which can be used to provide scores on four dimensions: independent living, relationships, mental health and senses. Alternatively, a single utility score may be computed by combining all of the scales except the illness scale; the score ranges from 1.00 (representing full HRQoL) to –0.04 (representing HRQoL states worse than death), with 0.00 representing death. The MCID appears to be 0.06. 151

The Dallas Pain Questionnaire (DPQ) is a 16-item instrument designed to evaluate the effect of chronic pain on four aspects of life: daily activities, work and leisure activities, anxiety/depression and social interest. Each item is scored on a scale whose extremities are marked 0% and 100%; although this scale appears to be continuous, for each item it is divided into five, six, seven or eight segments, each of which has a score. The item scores for each of the four aspects are added together and multiplied by a constant. They are then reported as a percentage, with lower scores indicating better quality of life. 152 No MCID was identified for the DPQ.

The Mini Mental State Examination (MMSE) is designed to measure cognitive status in adults. It is scored from 0 to 30, with higher scores indicating better cognitive condition. 153

The Short Form questionnaire-36 items (SF-36) is a questionnaire designed specifically to measure self-reported HRQoL; it has been proposed as the most appropriate measure of generic health status for use in people with spinal disorders. 149 It contains 36 questions which measure functional status, well-being and overall health in eight dimensions (physical functioning, role limitations owing to physical health, bodily pain, general health perceptions, vitality, social functioning, role limitations owing to emotional health, and mental health); these eight dimensions may be aggregated to produce summary measures of physical health [the physical component score (PCS)] and mental health [the mental component score (MCS)]. Results are presented on a scale of 0 to 100, with higher scores indicating better health, and may be transformed to take into account population norms. 154 Copay et al. 155 have suggested a MCID of 4.9 points specifically for the PCS, while Angst et al. 156 have suggested a MCID for improvement of 2.0 in the PCS, 7.8 in the bodily pain subscale, and 3.3 in the physical function subscale. Wiebe et al. 157 have suggested a MCID of 3.0 for the PCS and 4.6 for the MCS. No MCID has been identified for the overall SF-36 utility score.

Some of the studies included in this review only used the PCS, ignoring the scales for vitality, social functioning, role limitations owing to emotional health and mental health, thus undermining the value of using a generic measure of quality of life.

In addition to the generic quality of life measures, one measure, the Quality of Life Questionnaire of the European Foundation for Osteoporosis (QUALEFFO), has been developed and validated specifically for use in clinical trials in patients with vertebral osteoporosis. The QUALEFFO-41 has 41 questions relating to five domains: pain, physical function, social function, general health and mental health. 158 The domain scores are presented on a scale of 0 to 100, where 0 corresponds to the best HRQoL and 100 to the worst HRQoL. 159 QUALEFFO-41 has been shown to discriminate better between patients with and without vertebral fractures than the EQ-5D. 159 No MCID is known to have been suggested for the QUALEFFO. There is also a shorter version: the QUALEFFO-31. 160

Back-specific functional status/mobility

Measures of functional status assess the ability to perform specific tasks: this ability may have only a weak relationship with the reported level of pain. 161 However, for many patients, the greatest problem caused by a VCF is the limitation of activity rather than pain per se, and therefore functional status is a more clinically meaningful outcome than pain status. Moreover, it has been argued that the Roland–Morris Disability Questionnaire (RDQ), one of the most commonly used back-specific scales, is more useful than self-reported pain for assessing the impact of vertebral fractures on the patient’s daily life because it is more objective: self-reported pain is influenced by the patient’s perception and tolerance of pain, and relates only loosely to functional limitation. 162 This argument also applies to other measures of disability.

Furlan et al. 126 differentiate between back-specific functional status, as measured by items such as the RDQ, and disability, which can be assessed in terms of factors such as the ability to perform activities of daily living, work absenteeism, etc.; they consider both to be important patient-centred outcomes, along with symptoms (e.g. pain), perception of overall improvement, satisfaction with treatment, and well-being (e.g. quality of life measured with the SF-36, etc.). The studies included in this review use a number of different measures of functional status. These are detailed in alphabetical order, preceding a selection of direct observer-assessed measures of physical function.

The Barthel Index is a 10-item scale designed to evaluate the observer-assessed ability of a patient with a neuromuscular or musculoskeletal disorder to care for him or herself. The items assess independence in key activities of daily living (feeding, transferring from wheelchair to bed and back, grooming, toilet use, bathing, mobility on a level surface, ascending and descending stairs, dressing, faecal continence and urinary continence). It was originally scored from 0 to 100, with lower scores indicating greater disability,163 but may also be scored from 0 to 20, and in either case lower scores indicate greater disability.

A change of 1 point in the 0–20 Barthel Index (5 points in the original 0–100 version) represents a change in level of dependency in any of the key activities, and is therefore likely to be clinically meaningful. 57

The Oswestry Disability Index (ODI) was designed to measure patient-reported disruption to activities of daily living attributed to back pain. It comprises 10 dimensions [pain intensity, personal care, lifting, walking, sitting, standing, sleeping, sex life (if applicable), social life and travelling], each containing six response categories scored from 0 (indicating no disruption) to 5 (the worst state). The score is reported as a percentage of the maximum possible score (i.e. on a scale of 0–100). The weighted mean score for ‘normal’ populations is 10.19. 164

It has been suggested that, for the ODI, the MCID should be 4 points in relation to mean scores between groups but 15 points in patients before and after surgery. 164 More recently, Ostelo et al. 119 have suggested a MCID of 10 points in the transformed (0–100) scale, or 30% improvement from baseline, while Lauridsen et al. 165 suggest a MCID of 11 points in all patients, and 8 points in patients attending hospital back pain clinics.

Both the ODI and the RDQ are easy to use, reliable and valid. Because of floor and ceiling effects, the ODI is recommended for use with patients who are likely to have persistent severe disability, and the RDQ with patients who are likely to have relatively little disability, but for most patient groups both instruments appear to function satisfactorily in groups with severe disability. 164 However, the modified, 23-item version of the RDQ166 has been shown to be more responsive than the ODI in patients with low back pain only, whereas the ODI appeared slightly more responsive in patients with leg pain and/or low back pain. 165 Moreover, self-reported disability measures such as the RDQ have been shown to display only modest correlation with direct measures of physical function. 164

The RDQ was designed to assess physical disability due to low back pain. 164 It assesses self-reported functional status in eight dimensions (physical activities, housework, mobility, dressing, getting help, appetite, irritability and pain) by measuring 24 activity limitations. 167 Scores range from 0 (no disability) to 24 (maximum disability). 164 The original RDQ contained a six-point pain rating scale. However, this has now been excluded with the recommendation that the SF-36 pain scale be used for this purpose. 164 The modified RDQ (RDQ-23) contains 23 questions, some of which differ from those in the original questionnaire; it is scored from 0 to 23. 152

The MCID for the RDQ varies according to the level of disability of the patients, from 1 to 2 points in patients with little disability to 7–8 points in patients reporting high levels of disability, and 5 points in uncategorised patients; Ostelo et al. 119 suggest 5 points, or 30% improvement from baseline. Lauridsen et al. 165 suggest an overall MCID of 5 points, but recommend that 2 points should be used in populations attending hospital back pain clinics. However, Roland and Fairbank164 recommend that, in clinical trials which use the modified, 23-item version of the RDQ, a MCID of 2–3 points should be used for sample size calculations.

The Study of Osteoporotic Fractures-Activities of Daily Living (SOF-ADL) scale consists of six questions relating to activities undertaken in a typical day. The scale ranges from 0 to 18, with higher scores indicating more back-related disability. 153

Direct observer-assessed measures of physical function include:

• the timed ‘up and go’ test, which assesses functional mobility by measuring the time in seconds required to rise from a standard armchair, walk 3 metres, turn round, return to the chair and sit down again128

• the tandem test, which assesses balance by measuring the time for which the patient can stand in three different positions168

• the repeated chair test, which tests muscle power by asking the patient to rise from, and return to, sitting as many times as possible in 30 seconds; higher scores indicate better health status. 153

The included studies also assessed functional status in terms of the use of aids and appliances, days of bed rest or reduced activity, and perceived recovery.

Pain/analgesic use

The recommended measure of global pain severity is the bodily pain subscale of the SF-36, a two-item scale which measures pain intensity on six levels (from none to very severe) and interference with activities on five levels (from not at all to extremely). 149 This measure is recommended not least because of the availability of normative data. 169 An absolute cut-off value has been suggested for the MCID of 3 points overall or 2 points in populations attending hospital back pain clinics. 165

However, many of the included studies used either:

• a visual analogue scale (VAS), whereby patients mark the point which best represents their pain on a line, usually 10 cm long, whose ends bear labels describing the extremes of pain intensity (e.g. ‘no pain’ and ‘worst imaginable pain’); or

• a numeric rating scale, whereby patients rate the intensity of their pain on a scale of 0 to 10 (11-point scale) or 0 to 100 (101-point scale), where 0 represents no pain and 10 or 100 the worst possible pain. 169

The VAS, which is formatted without numbers, represents a continuous range of values, while the numeric rating scale is formatted using whole numbers to form a segmented scale. 170

Because it is usually measured in millimetres, and can therefore be regarded as having 101 response categories, a 10 cm VAS is potentially more sensitive than an 11-point numeric rating scale. However, it has been shown to be more difficult to understand than other measures of pain intensity, especially for people at risk of cognitive difficulties (e.g. some elderly individuals, or people taking high doses of opioid analgesics). 169 Moreover, the VAS was developed for assessing chronic pain and is less reliable in the immediate postoperative period, when any single VAS score may have an imprecision of ±20 mm (20%). 171 Numeric rating scales are easier to use, and their sensitivity and validity has also been demonstrated. Both Bolton and Wilkinson172 and Grotle et al. 173 suggest that the numeric rating scale may be more responsive than the VAS. However, unlike the VAS, numeric rating scales have not been demonstrated to have ratio qualities (i.e. a change in pain score from 9.0 to 6.0 cannot be assumed to represent a 33% decrease in perceived pain). 169

Some of the included studies (Blasco et al.,127 Farrokhi et al.,135 Liu et al.,139 Rousing et al. 128 and VERTOS II174) stated that they used a VAS scale; others (Buchbinder, et al. 148 FREE,137 INVEST147 and VERTOS136) specified that they used an 11-point numeric rating scale, although in the FREE137 and VERTOS136 studies this was called a VAS.

As the distribution of scores from VAS and numeric rating scales is not normal, non-parametric statistical analyses are appropriate. 57 Moreover, because initial and subsequent pain ratings on the VAS tend to be correlated, any between-group comparison should compare changes in scores from baseline rather than simply differences in the final post-treatment score. 57

Ostelo et al. 119 have proposed an absolute cut-off value for the MCID of 15 on a 100-unit VAS, with a relative cut-off value a 30% improvement from baseline. However, DeLoach et al. 171 suggest that, because of the imprecision found in the immediate postoperative period, the MCID in that period should be 20 out of 100 units. Ostelo et al. 119 also proposed that, when an 11-point numeric rating scale is used for low back pain, the absolute cut-off value for the MCID should be 2 points, again with a relative cut-off value of a 30% improvement from baseline. However, Copay et al. 155 suggest a MCID of 1.2 points for back pain.

The INVEST study also used a modified version of the Deyo–Patrick pain frequency and bothersomeness scale. 147 The original scale measured the frequency with which patients experience pain, and how pain impacts on their daily life, on a scale of 0 to 6, where 6 indicates the highest impact. 166 The INVEST study used a scale of 0 to 4, with higher scores again indicating more severe pain. 103

While the technology assessment protocol did not specify analgesic use as an outcome, it has been included because some of the included studies used analgesic use as a proxy for pain relief. This may be measured either quantitatively (i.e. amount of analgesia used) or qualitatively (i.e. type of analgesia used). Most of the included studies used a qualitative approach, grouping analgesics into categories such as non-opioid, weak opioid and strong opioid. Doidge et al. 57 note that these distinctions are clinically important because the risk profiles of the different categories of analgesic differ substantially.

Vertebral body height and angular deformity

Risk of selection bias

All the included studies were described as randomised. However, as Blasco et al. 127 stated only that ‘randomisation was done with a previously defined randomised computer list’, there was lack of clarity about the method of both assignment to treatment groups and concealment of allocation. Liu et al. 139 provided no information regarding concealment of allocation.

In multicentre trials of interventional procedures, stratification of randomisation by treatment centre is important to avoid potential imbalances associated with differences in techniques and skills. Two of the five multicentre studies (Buchbinder et al. 102 and INVEST103) specified that randomisation was stratified by treatment centre; in a third (FREE137), although randomisation was stratified, treatment centre was not one of the variables. In VERTOS136 and VERTOS II,17 randomisation is not said to have been stratified. Of the remaining four studies, that by Blasco et al. 127 was definitely a single-centre study, and those by Liu et al. 139 and Rousing et al. 128 were probably also single centre. In the study by Farrokhi et al. ,135 although two hospitals appear to have been involved, the same surgeon seems to have undertaken the procedure in both. 197 Thus, the risk of bias owing to differences in techniques and skills appears to be higher in FREE, VERTOS and VERTOS II than in the remaining studies.

Risk of performance bias

Few studies appear to be at risk of bias associated with co-interventions as they offered the same oral analgesics to both the treatment and control groups. In the study by Blasco et al. ,127 the methods section suggests that rescue therapy by intrathecal infusion (25 µg fentanyl and 1.5 mg bupivacaine) was only offered to patients in the control group in whom drug therapy had proved ineffective (VAS > 7) or intolerable. If there was then no improvement in pain, conservative treatment was deemed to have failed and the patient was considered for PVP. However, it is clear from the results section that this rescue therapy was also made available to patients who had received PVP. 127 The risk of performance bias was unclear in the study by Rousing et al. :128 this did not explicitly define conservative treatment or detail the steps taken to avoid co-interventions, but appeared to offer brace treatment to the control group but not the PVP group.

Some of the studies were at risk of bias because of crossover – that is to say patients who were randomised to one intervention receiving the other intervention. Crossover is generally more common in unblinded than in blinded studies because, in unblinded studies, the patients in the control group are aware that they have not received the experimental intervention. Three of the included studies appear to be free of crossover: it was not allowed in the studies by Buchbinder et al. 102 and Rousing et al. ,128 and was presumably impossible in Liu et al. ’s139 comparison of PVP with BKP. In the remaining six studies, patients randomised to one treatment might choose, after a minimum period of time, to receive the other treatment; in the FREE study, they were then considered to have withdrawn from the study, but were included in the intention to treat analysis at 1 year. 137 In the blinded INVEST study,103 patients were informed at recruitment that they would be allowed to cross over to the other procedure 1 month or more after the intervention if adequate pain relief was not achieved; specific numerical pain thresholds were not used to determine the adequacy of pain relief and therefore eligibility to cross over. 103 Although crossover was permitted, blinding was maintained for the full year. 147 In the unblinded studies, crossover was reported to be in one direction only, from conservative treatment to the intervention; however, in the Blasco et al. ,127 FREE137 and VERTOS II17 studies, some patients allocated to PVP or BKP did not receive the intervention for various reasons, and the investigators did not appear to adjust for non-treatment. The blinded INVEST study103 reported crossover from PVP to the OPLA as well as from the sham procedure to PVP: by 3 months, 8 out of 68 (12%) of patients allocated to PVP had crossed over, compared with 27 out of 63 (43%) allocated to OPLA103 ( Table 6 ).

Several studies which were theoretically at risk of bias owing to crossover attempted to avoid such bias by reporting results only up to the point at which crossover was permitted. Thus, the INVEST study,103 which permitted crossover after 1 month, reported comparative results only up to that point; 2 out of 42 patients (5%) appear to have crossed over from OPLA to PVP before 1 month. 103 The VERTOS study,136 which permitted crossover at 2 weeks, was stopped prematurely at that point because 88% of the control group (14/16) requested PVP. 136 In the study by Blasco et al. ,127 7 out of 61 patients allocated to conservative therapy (11%) underwent PVP,127 but the date at which this occurred was not reported. Farrokhi et al. 135 permitted crossover after 1 month; 10 of the 42 control patients (24%) had undergone PVP by 1 year, and a further 10 apparently underwent PVP after 1 year. 135 In the FREE study,137 14 out of 151 patients in the control group (9%) underwent BKP, nine of them (6%) before 1 month. 137 In VERTOS II,17 5 out of 101 patients withdrew from the control group before treatment because they wanted PVP; because they withdrew consent, the vertebroplasty procedure could not be documented and analysed in those patients. A further 10 control patients (10%) requested PVP at various points during the study. 17 Thus, the studies with the highest rates of crossover (INVEST103 and VERTOS136) report comparative results only for the period preceding any substantial crossover, and the 1-month data from the study by Farrokhi et al. 135 are also unaffected. However, there is some potential for bias owing to crossover in the study by Blasco et al. 127 and the FREE137 and VERTOS II17 studies.

Only two studies (Buchbinder102 and INVEST103) sought to avoid bias by blinding patients to their treatment allocation; both used an OPLA to do so. In the INVEST study, the success of blinding was evaluated: at 14 days, 51% of patients in the PVP group and 63% in the control group correctly guessed their allocation, but in both cases their degree of confidence in the accuracy of their guess was only moderate. 103 Buchbinder et al. planned to evaluate the success of blinding at the end of the study148 but do not appear to have reported the results. In the study by Liu et al. , patient blinding was presumably feasible, as the patients received PVP or BKP under general anaesthesia, but, as patient blinding was not mentioned, it seems unlikely that it was done, particularly as it was stated that some outcomes were assessed by blinded assessors. 139 The remaining studies made no attempt to blind patients to their treatment allocation. Consequently, as many of the outcomes were patient-reported and subjective in nature, the risk of bias associated with lack of blinding in these studies is substantial.

Risk of detection bias

Because of the radio-opaque nature of the cement used for PVP and BKP, it is impossible to blind the assessors of radiographic outcomes (vertebral body height, kyphotic angle and incident fracture) to treatment allocation. However, there is no reason why blinded assessors should not have been used to collect data relating to other outcomes, yet only four studies (Buchbinder et al. ,102 Farrokhi et al. ,135 INVEST103 and Liu et al. 139) stated that they used blinded outcome assessors for at least some outcomes. Three studies (Blasco et al. 127, FREE137 and VERTOS II186) stated that the radiopacity of the bone cement made it impossible to blind outcome assessors, and appeared to make no attempt to use blinded assessors for non-radiological outcomes. There appeared to be no blinding of outcome assessors in the Rousing et al. 128 and VERTOS136 studies. As many of the non-radiological outcomes were subjective patient-reported outcomes, which may be modified by contact with non-blinded outcome assessors, such data are at risk of bias in all except the Buchbinder et al. 102 and INVEST103 studies, and possibly also that by Liu et al. 139

In the majority of studies, outcomes appeared to be assessed at comparable times in both groups. However, in the study by Liu et al. , the timing of the assessment of some outcomes was not clear. 139 In VERTOS II, the timing of assessment of both baseline characteristics and subsequent outcomes appears not to be comparable because baseline was said to be the day of randomisation for the control group but the day of PVP for the intervention group; moreover, PVP was said to have been performed a mean of 9.4 (SD 8.1) days after randomisation. 17 This has two major consequences:

• Although the inclusion criteria stipulate that participants should have had back pain for no more than 6 weeks, many patients in the PVP group would have undergone the intervention more than 6 weeks after pain onset, and thus would have subacute rather acute VCFs, whereas all of the control group would have had acute fractures at baseline. Consequently, Doidge et al. 57 suggest that between-group differences in baseline variables such as the EQ-5D, which Klazen et al. 17 ascribe to chance, may in fact be due to differences in fracture acuity.

• In the control arm of VERTOS II, the mean pain score fell by 1.9 points (25%) during the first week following randomisation. 17 A comparable reduction might presumably be expected in the PVP group between randomisation and the assessment of baseline characteristics on the day of the intervention. However, as the ‘baseline’ pain score was 7.8 in the PVP group compared with 7.54 in the control group, if such a reduction occurred, it implies both a substantial disparity between groups at randomisation and an implausibly high mean pain score in the PVP group at that point in time.

There is thus considerable lack of clarity in relation to the timing of assessments in VERTOS II, and clarification has been sought, but not received, from the study first author.

Risk of attrition bias

Five studies (Buchbinder et al. ,102 Farrokhi et al. ,135 INVEST,103 Rousing et al. 128 and VERTOS II17) followed up at least 80% of participants. Liu et al. 139 made no reference to attrition: this may be because there was none, but this is not specified. In the Blasco et al. 127 and FREE137 studies, only 76% and 78%, respectively, completed follow-up at 12 months; in the FREE study, there was a marked disparity between treatment groups, with 83% in the BKP group completing follow-up at 12 months compared with 74% in the control group,137 whereas in the study by Blasco et al. follow-up was higher in the control group (79% vs. 73%). 127 In the VERTOS study, only 74% overall completed 2 weeks’ follow-up. The data are poorly presented, making a full comparison of drop-out rates in the two treatment groups impossible; six patients are said to have withdrawn from the control group because they wanted PVP, and two patients from the PVP group because they wanted the control intervention, but details are not given of the treatment allocation of the four patients who refused to complete questionnaires at 2 weeks, nor are their outcomes reported at 1 day. 136

All studies except Liu et al. 139 gave information relating to the reasons for withdrawal. While the absence of such information in Liu et al.’s study may further suggest that there were no withdrawals, this was not explicitly stated.

With the exception of VERTOS,136 all studies specified how many patients were randomised to each group. Moreover, all but Liu et al. 139 stated clearly how many were included in the final analysis, and it is possible that Liu et al. 139 did not provide this information because there were no withdrawals. 139 The majority of studies, including all the studies which reported crossover, reported using intention-to-treat analyses.

Risk of reporting bias

Only three studies (Blasco et al.,127 Farrokhi et al. 135 and VERTOS II17) appeared to be free of selective reporting. The Buchbinder et al.,102 FREE137 and INVEST103 studies reported most but not all of the clinical outcomes specified in the study protocol. It is understood that the longer-term outcomes from the Buchbinder et al. study (presumably including the incidence of new vertebral fractures at 12 and 24 months), although not yet reported, are to be published; however, this does not explain the failure to report the results of the timed ‘up and go’ test and the patients’ perceptions of fatigue and overall health. The FREE study did not report functional outcome and the results of objective functionality tests or data relating to vertebral body height (for details, see Appendix 9 , Table 101 ); the last of those omissions is the most surprising, given that one of the particular merits of BKP is said to be its effect on VBH. The INVEST study did not report adjacent fractures at 12 months or implant-related inflammation. No study protocol could be found for the Liu et al.,139 Rousing et al. 131 or VERTOS136 studies, and therefore the risk of reporting bias in these studies is not clear.

External validity

The external validity of the included trials is summarised in Figure 3 . Most of the included studies specified their eligibility criteria, thus enabling assessment of the nature of their patient populations. However, many reported that a substantial proportion of patients declined to participate, and only Farrokhi et al. 135 specifically stated that the rate of refusal to participate was low (see Table 7 ). It is not always clear whether patients refused to participate before or after they were found to meet the study inclusion criteria. Consequently, the figures included in Table 7 are presented as percentages of the total number of potential patients who were said to have been screened for each study, and not of the (lower) number remaining following subtraction of those who were subsequently excluded because they did not meet the inclusion criteria or for other reasons; the figures may thus be regarded as conservative. As refusal to participate is a patient’s decision rather than one made by the study investigators, such a decision made prior to randomisation may be expected to affect both treatment groups equally and seems unlikely to affect study validity, although it may limit generalisability if the decision to participate is influenced by symptom severity. However, in the INVEST study, a comparison of data relating to eligible patients at the lead site who did and did not enrol found no significant differences in age, proportion of women or RDQ score; data relating to pain were collected differently in the two groups and were therefore not directly comparable. The authors therefore suggested that the results of the INVEST study should be generalisable to all patients who would have been eligible for enrolment in that study. 197

Although most of the included studies provided an explicit description of the interventions, Rousing et al. provided only a relatively cursory description of the control treatment. 128 Traditionally, PVP has been performed by radiologists and BKP by surgeons. 198 However, few of the included studies provided adequate details of the clinical background and specific procedure-related training of the clinician who performed PVP or BKP, or of their relevant experience (i.e. the number of procedures they had completed before the study), although such information is important for interpreting study results. Studies involving inexperienced clinicians and centres in the early stages of introducing PVP or BKP may include ‘learning curve’ data, whereas studies involving more experienced clinicians and centres may have more favourable results. The information provided was judged unclear if only the specialism (e.g. radiology or neurosurgery) was reported; it was only judged adequate if details were also given of the specific training in PVP or BKP which the operators had received or the number of such procedures which they had previously performed. Thus, Buchbinder et al. 102 specified that PVP was performed by experienced interventional radiologists who had undertaken formal training in vertebroplasty, had appropriate certification, were actively performing the procedure, and all adhered strictly to a detailed, standardised protocol, while in the INVEST study103 PVP was said to be performed by highly experienced practitioners who had performed a mean of approximately 250 procedures (range 50–800). Blasco et al. ,127 Farrokhi et al. 135 and Rousing et al. 128 provided information relating to the clinicians’ specialism (respectively experienced neurointerventional radiologists, a neurosurgeon, and orthopaedic surgeons specialising in spine surgery) but did not specify their training or level of experience of PVP. The remaining studies (FREE,137 Liu et al. ,139 VERTOS136 and VERTOS II17) provided no relevant information.

All studies used relevant outcome measures, and all but Liu specified that they used valid instruments. All assessed short-term outcomes. However, some studies did not either measure or report long-term outcomes; as noted earlier, 1-year assessments were planned in INVEST103 and VERTOS136 but because of crossover, VERTOS was stopped at 2 weeks,136 while INVEST followed patients up for 1 year but reported outcome data only at 1 month. 103 Buchbinder et al. 102 followed patients up for 2 years, and the 1- and 2-year data are currently being prepared for publication, as is a separate paper on radiological outcomes (Rachelle Buchbinder, Monash University, 2012, personal communication).

Most studies provided an adequate description of adverse events.

Precision

Only three of the included studies (Blasco et al. ,127 FREE137 and INVEST103) appeared to be adequately powered for at least their primary outcome: pain as measured on an 11-point scale (Blasco et al. and INVEST) and change from baseline to 1 month in the SF-36 PCS scale (FREE). However, because of difficulties with recruitment, the power of the INVEST study was reduced from a power of more than 80% to detect a 2.5-point difference between groups on the RDQ score and a 1.0-point difference on an 11-point pain scale to a power of more than 80% to detect a 3.0-point difference on the RDQ score and a 1.5-point difference on the pain scale. 103 Because most studies were underpowered for most outcomes, the absence of a statistically significant difference between treatment groups does not necessarily mean that such a difference would not be found in a larger study.

Almost all studies except that by Blasco et al. 127 published point estimates and measures of variability; Blasco kindly supplied additional data in that format (Jordi Blasco Andaluz, Hospital Clinic de Barcelona, Spain, 2012, personal communication).

FIGURE 3.

External validity and precision summary: review authors’ judgements about each included study (+, good generalisability/precision; –, poor generalisability/precision; ?, unclear generalisability/precision).

Health-related quality of life

European Quality of Life-5 Dimensions

Five studies (Buchbinder et al. ,102 FREE,137 INVEST,103 Rousing et al. 128 and VERTOS II17) collected EQ-5D data. However, two studies did not collect relevant data from all participants, although in both cases non-collection of EQ-5D data does not appear to be related to patient characteristics. Buchbinder et al. only added this outcome to their protocol in June 2005 to allow comparison with the INVEST trial study, and therefore EQ-5D scores were available for only 30 out of 38 participants (79%) in the intervention group and 29 out of 40 (73%) in the control group. 102 Similarly, Rousing et al. collected EQ-5D data only from November 2004, when a PhD study was affiliated to the trial: thus, scores were available for only 15 out of 26 participants (58%) in the intervention group and 17 out of 24 (71%) in the control group. 128 VERTOS II collected EQ-5D data throughout the study and reported baseline values, but the investigators did not report follow-up values, although they used them to estimate quality-adjusted life-years (QALYs). 17

The two blinded RCTs (Buchbinder et al. 102 and INVEST103) found no significant difference between PVP and conservative treatment in terms of short- or medium-term outcomes (see Appendix 10 , Table 105 ). As Doidge et al. note, the CIs include effects which might favour either group, suggesting that the studies were underpowered to detect clinically important differences in this outcome. 57 However, when 1-month data from the Buchbinder and INVEST studies were combined in a meta-analysis of IPD,110 the result was not statistically significant (see Appendix 10 , Table 107 ) and, because the MCID is 0.08,150 the CI for the pooled data only just includes the possibility of a clinically important difference favouring PVP. In the study by Rousing et al. , the changes from baseline at 3 and 12 months favour conservative treatment and suggest that the difference between groups is clinically important.

The FREE study found statistically significant differences in outcomes, favouring BKP over non-surgical management, at 1, 12 and 24 months (see Appendix 10 , Tables 105 and 106 ). However, although at each time point the point estimate is greater than the MCID, the CIs at 3, 6, 12 and 24 months include the possibility of clinically unimportant effects.

Figures 4 , 5 , 6 and 7 graphically represent these change in EQ-5D observed in the Buchbinder et al.,102 FREE,137 INVEST103 and Rousing et al. 131 trials respectively.

FIGURE 4.

European Quality of Life-5 Dimensions data recorded in Buchbinder et al. 102

FIGURE 5.

European Quality of Life-5 Dimensions data recorded in the FREE trial. 137

FIGURE 6.

European Quality of Life-5 Dimensions data recorded in the INVEST trial. 103

FIGURE 7.

European Quality of Life-5 Dimensions data recorded in the Rousing trial. 131

Back-specific functional status/mobility

All of the studies, except that by Liu et al., 139 reported some measure of back-specific functional status or mobility.

Pain/analgesic use

Pain

Only one study, the FREE study,137 reported pain using the recommended measure of global pain severity, the bodily pain subscale of the SF-36,169 in which higher scores represent better health. The only result which has been published from this study using this measure is the difference between treatment groups in average improvement over a period of 12 months: this was 9.2 points greater in the BKP group than in the control group (95% CI 3.9 to 14.6, p = 0.0008). 137 Fuller confidential data presented in supplementary document 8 of the manufacturer’s submission35 indicate that (academic-in-confidence information has been removed) (see Appendix 10 , Table 112 ). (Academic-in-confidence information has been removed.)

All nine studies reported pain measured on either a numeric rating scale or a VAS, with higher scores indicating more severe pain. Farrokhi et al. 135 and INVEST103 asked patients to report their average pain over the previous 24 hours, while Buchbinder et al. 102 and FREE137 asked them to do so over the previous week, and the remaining studies did not specify the time period. However, empirical data indicate that broadly comparable results are obtained regardless of whether patients are asked to report average pain over the previous 24 hours or the previous week. 199 Academic in confidence data were provided for the Buchbinder et al. 102 RCT (Margaret Staples, Monash University, 2012, personal communication) for the VAS scores at 12 and 24 months.

As noted in Chapter 2 (see Decision problem), the VAS is less responsive than the numeric rating scale; this is presumably the reason why Doidge et al. have suggested that data collected by the two methods should not be combined in a meta-analysis. 57 The majority of the included studies (Buchbinder et al. ,102 Farrokhi et al. ,135 FREE,137 INVEST,103 Liu et al. ,139 VERTOS136 and VERTOS II17) clearly used a numeric rating scale; although some termed it a VAS, they also referred to it as a 10-point scale. It is not wholly clear whether or not Blasco et al. 127 actually used a VAS, although they claimed to do so. Rousing et al. 131 specified that they used a 10-cm VAS,128 and presumably did so at most time points, but they clearly used a numeric scale in a supplementary telephone interview in which, after all but three had completed 12 months’ follow-up, patients were asked to rate their back pain 1 month after discharge from hospital on a scale of 0–10. 131 As Doidge et al. have pointed out,57 because these data were collected almost 1 year after the event, they are at high risk of recall bias.

Farrokhi et al. ,135 FREE,137 Rousing et al. 131 and VERTOS II17 found statistically significant differences between groups in short- and medium-term changes from baseline in pain following PVP or BKP; FREE and VERTOS II also found statistically significant long-term differences between groups (see Appendix 10 , Tables 124–126 ). However, the double-blinded studies (Buchbinder et al. 102 and INVEST103), and the small VERTOS136 study, found no significant differences between treatment groups, while in the study by Blasco et al. 127 statistical significance in change from baseline was only reported at 2 months, when the result favoured PVP. There appears to have been no significant difference between treatment groups in terms of change from baseline at 12 months, and Blasco et al. attribute the similar prevalence of moderate and/or severe residual pain to the more frequent use of rescue therapy in the control group and the higher number of new clinical fractures associated with PVP in the intervention group. 127 Moreover, the favourable result reported by Rousing et al. 128 at 1 month (see Appendix 10 , Table 125 ) is unreliable because of the high risk of recall bias discussed above. Liu et al. 139 found no significant differences between PVP and BKP, but the study does not appear to have been powered to do so.

Meta-analysis of IPD from the Buchbinder et al. 102 and INVEST103 studies again found no significant difference at 1 month in terms of mean pain scores (see Appendix 10 , Table 127 ).

A comparison of longitudinal trends in pain reduction between the differently treated groups proves instructive. Figures 8 to 13 graphically represent these trends for PVP, BKP, OPLA, and conservative treatment respectively. Graphs for the longitudinal pain changes in individual trials are also included in Appendix 12 . Among the cohorts treated with PVP and BKP, there is a rapid post-procedural reduction in pain which appears to stabilise at approximately 1 month. The OPLA-treated cohorts reveal a somewhat similar pattern: there is a rapid reduction in pain, which seems to stabilise at 1 month. In contrast to PVP and BKP, however, there appears to be a small, temporary worsening of pain between 1 day and 1 month, at which point pain once again reduces and stabilises. A rather different pattern emerges with respect to those treated with OPM. There is no dramatic initial drop in pain; rather, there is a more gradual reduction until approximately 3 months, at which point the pain level stabilises and becomes comparable to that of those treated with PVP.

FIGURE 8.

Longitudinal pain reduction trends in vertebroplasty without AiC data.

FIGURE 9.

Longitudinal pain reduction trends in vertebroplasty with AiC data. (Academic-in-confidence information has been removed.)

FIGURE 10.

Longitudinal pain reduction trends in BKP.

FIGURE 11.

Longitudinal pain reduction trends in OPLA excluding AiC data.

FIGURE 12.

Longitudinal pain reduction trends in OPLA including AiC data. (Academic-in-confidence information has been removed.)

FIGURE 13.

Longitudinal pain reduction trends in OPM.

The gradual reduction in pain seen in conservatively treated patients coheres with a regression to the mean as would be expected from the natural history of healing in osteoporotic VCFs. The patterns seen in the PVP and OPLA groups, on the other hand, pose some more interesting interpretive questions. Whitehouse has suggested that the initial ‘dip’ seen in the OPLA cohorts represents a strong initial placebo effect before regression to the mean, while the early and sustained reductions in the PVP cohorts is suggestive of specific mechanisms of efficacy. 95 However, owing to the truncated line from INVEST, interpretations should be made with caution.

(Academic-in-confidence information has been removed.)

If, as indicated in Decision problem, a difference between groups of 2 or more points indicates a clinically meaningful difference, then most of the short- to medium-term results which are statistically significant also appear to be clinically meaningful. However, although FREE137 and VERTOS II17 both reported statistically significant longer-term results, in the FREE study these results did not appear to be, and in VERTOS II the 95% CI included the possibility that they were not, clinically meaningful (see Appendix 10 , Table 126 ).

The INVEST study also stated that 64% of patients randomised to PVP and 48% of those randomised to OPLA reported a clinically meaningful improvement in pain (i.e. a decrease of 30% or more) at 1 month (p = 0.06). 103 It has been suggested that this trend towards favouring PVP might have achieved statistical significance if the trial had recruited more participants, as was originally planned. 200 However, when these data were combined with those from the Buchbinder et al. 102 study in a meta-analysis of IPD, no significant difference at the 5% level was found in the proportion showing a clinically meaningful improvement in pain at 1 month, whether this was defined as a decrease in pain of at least 3 units or of at least 30% ( Table 11 ).

TABLE 11 - Number of patients in the Buchbinder and INVEST studies showing improvement in pain scores at 1 month: data from Staples et al. 2011110

NS, not significant.

Outcome	PVP	Control	Relative risk (95% CI)	p-value
Improvement in pain of ≥ 3 units	55/102 (53.9%)	43/99 (43.4%)	1.3 (0.8 to 1.9)	NS
Improvement in pain of ≥ 30%	61/102 (59.8%)	45/99 (45.5%)	1.3 (1.0 to 1.8)	NS

In VERTOS II, survival analysis indicated that significant pain relief (apparently defined as a decrease from baseline in pain score of 3 points or more) was achieved earlier, and in more patients, after PVP than with conservative treatment [29.7 days (95% CI 11.45 days to 47.97 days) vs. 115.6 days (95% CI 85.87 days to 145.40 days) (χ² = 55.6, p < 0.0001)]. 17

Blasco et al. ,127 Buchbinder et al. 102 and VERTOS136 also reported pain outcomes in terms of QUALEFFO pain scores. The reported figures are not directly comparable as they appear to use different scales, although this is poorly reported: Blasco appears to report the domain score (scored from 0 to 5) whereas Buchbinder reports pain scores on a scale of 0 to 100, and it is not clear what potential range of scores is represented by the VERTOS data. Blasco et al. and Buchbinder et al. found no statistically significant difference between the groups; the significance of the VERTOS results unfortunately could not be calculated (see Appendix 10 , Table 128 ).

The INVEST study also reported on the frequency with which participants experienced pain and the impact of pain on their daily lives, both measured on a scale of 0–4. In both groups, pain frequency and pain bothersomeness decreased between baseline and 1 month; however, although the point estimates favoured the intervention, the results were not statistically significant103 (see Appendix 10 , Table 129 ). Moreover, as Doidge et al. note, the CIs did not include a 1-unit effect (the smallest possible threshold of clinical importance) in either direction. 57

Buchbinder et al. 102 collected data on perceived pain: this was classified as ‘better’ if the patient indicated that it was moderately or a great deal better than before the intervention, and ‘worse’ if they reported that it was moderately or a great deal worse. They found no statistically significant between-group differences in the proportion of patients in these categories at any time point102 (see Appendix 10 , Table 130 ).

Vertebral body height and angular deformity

Four studies (Blasco et al. ,127 Farrokhi et al. ,135 FREE137 and Liu et al. 139) reported changes in VBH and/or angular deformity. However, their results are not necessarily comparable as it is not clear whether or not they used the same methods of measuring vertebral height. Farrokhi et al. 135 calculated the mean VBH by taking the mean of the height of the anterior wall plus the height of the posterior wall, while Blasco et al. 127 and Liu et al. 139 referred to the mean height without specifying how it was measured. Farrokhi et al. specified that they used the sagittal index to measure angular deformity135 whereas the FREE study137 and Liu et al. 139 used the kyphotic angle (see Chapter 2, Decision problem). It is not clear whether or not Liu et al. 139 measured postoperative VBH and angular deformity at 3 days or at 6 months. Because of these potential sources of heterogeneity, it did not seem appropriate to pool data relating to VBH or angular deformity.

Surprisingly, the FREE study did not report changes in VBH even though maintenance of VBH was one of its secondary outcome measures202 and is one of the respects in which BKP might be expected to provide additional benefit compared with PVP. However, although kyphotic angle was measured in both groups, the study protocol stated that VBH was to be measured only in patients undergoing BKP,202 thus making comparison with control subjects impossible.

Blasco et al. 127 found no significant difference between treatment groups in change in VBH from baseline at 12 months. By contrast, Farrokhi et al. 135 found that PVP was associated with significant improvements in mean VBH which were sustained throughout the first year but not thereafter, and with significant improvements in angular deformity which were sustained throughout the 36-month follow-up period (see Appendix 10 , Tables 134 and 135 ). They suggest that the significant differences from pre-treatment values in mean VBH and sagittal index seen at 1 week in the PVP group (p < 0.002 and < 0.011 respectively) but not in the control group (p = 0.22 and < 0.80 respectively) may be related either to the prone position used during PVP or to the high pressure produced within the vertebra by the injected cement, both of which can expand the vertebra and correct kyphotic deformity to some extent. 135

The FREE study only reported improvement from baseline in the kyphotic angle of the index fracture at 24 months, without reporting the absolute figures at either point in time. They reported a statistically significant result in favour of BKP;201 however, the clinical significance of this result is not clear. In the study by Liu et al. ,139 BKP was associated with significantly greater improvements in both postoperative VBH and angular deformity than was PVP (see Appendix 10 , Tables 134 and 135 ).

Progression of treated fractures

Only one study, VERTOS II, reported data relating to the progression of treated fractures during follow-up; in the control group, all vertebrae which showed bone oedema on baseline MRI were considered to be treated vertebrae. At last follow-up (mean 11.4 months, median 12.0 months, range 1–24 months), moderate or severe height loss was seen in 11 vertebrae in 11 out of 91 patients (12%) in the PVP group compared with 39 vertebrae in 35 out of 85 patients (41%) in the control group (p < 0.001)186 (see Appendix 10 , Table 136 ).

All-cause mortality

Six of the included studies reported all-cause mortality. 17,127,128,135–137 Liu et al. 139 made no reference to any deaths, thus implying that none occurred; however, this was not explicitly stated. None of the individual studies found any statistically significant differences in overall mortality between treatment groups (see Appendix 10 , Table 137 ). However, this is unsurprising, as they were not powered for this outcome. None of the reported deaths appear to be related to treatment: the patient in VERTOS II who died as a result of gastric bleeding had used morphine as his or her only analgesic. 17

Three studies (Blasco et al. ,127 Rousing et al. 131 and VERTOS II17) reported overall mortality at the same time point (12 months). Data from these studies were combined by meta-analysis; inclusion of data from other studies which reported mortality at different time points was not considered appropriate. Statistical significance was still not achieved when the data from these studies were pooled, although the point estimate favours PVP ( Figure 14 ).

FIGURE 14.

Overall mortality at 12 months. M–H, Mantel–Haenszel.

Symptomatic and asymptomatic cement leakage

Seven studies (Blasco et al. ,127 Buchbinder et al. ,102 Farrokhi et al. ,135 FREE,137 Rousing et al. ,128 VERTOS136 and VERTOS II17) reported cement leakages identified using imaging equipment. Four studies (Blasco et al. ,127 Buchbinder et al. ,102 Farrokhi et al. 135 and Rousing et al. 128) appear to have reported only leakages identified by fluoroscopy during the procedure, whereas two (VERTOS136 and VERTOS II17) performed CT immediately after PV to identify possible cement leakage or other local complications; this technique is likely to identify more leaks than fluoroscopy. The FREE study assessed cement extravasation using both intraoperative fluoroscopy and postoperative radiographs,137 thus increasing their likelihood of identifying leaks compared with the use of fluoroscopy alone (see Decision problem). All seven studies stated that they used PMMA cement; none referred specifically to high-viscosity cement and it is therefore assumed that low-viscosity cement was used in all studies.

For PVP, the number of treated vertebrae in which cement leakages were reported ranged from none in the small VERTOS136 study to 72% in VERTOS II;17 the pooled data suggest an incidence of 44% for PVP compared with 27% for BKP. However, this approach may conceal a relationship between the volume of cement injected and the likelihood of leakage, or between the sensitivity of the method of detection used and the detection rate: thus, the highest incidence is seen in VERTOS II,17 which also reports the highest mean volume of cement injected per vertebra, and which specifically scanned patients postoperatively using CT scanning, the most sensitive method of detection, to identify possible leakages ( Table 13 ).

The importance of cement leaks relates to their potential clinical sequelae. These may be immediate or delayed. Blasco et al. found that, although the cement leaks which they reported were not associated with immediate clinical complications, cement leakage into the inferior disk was associated with an increased risk of incident vertebral fracture [OR 7.17 (95% CI 1.69 to 69.30), p = 0.0008]. 127 Farrokhi reported 13 asymptomatic leaks (five into the discal space and eight into the paravertebral space) and one symptomatic leakage into the epidural space. The symptomatic leakage caused severe right lower-extremity pain and weakness but, following immediate decompression through a bilateral laminectomy and evacuation of bone cement, the patient could walk unassisted with no radicular pain after 2 months. 135 Rousing stated that none of the cement leaks caused neurological symptoms. 128 In VERTOS II, most leakages were discal or into segmental veins; none was into the spinal canal. All patients remained asymptomatic, even though fluoroscopy showed cement migration into the venous system towards the lungs in one patient; a follow-up chest CT after 1 year showed no perifocal inflammatory pulmonary changes in this patient. In this study, an asymptomatic cement deposition in a segmental pulmonary artery was also reported, presumably in another patient. 17 Fifty-four PVP patients subsequently underwent CT after a mean follow-up of 22 months (median 21 months, range 6–42 months). Although during the procedure the operators had not reported fluoroscopically visible cement migration towards the lungs in any of these patients, at follow-up 14 out of 54 (26%, 95% CI 16% to 39%) had pulmonary cement embolism visible on CT. The emboli varied in size between 1 mm and 12 mm and were randomly distributed in the periphery of the lungs; six patients had a single cement embolus, while the remaining eight had between 2–35 cement depositions randomly scattered in the peripheral portions of both lungs. All of the affected patients were asymptomatic. 193 In the FREE study, most leaks were endplate or discal leakages, with one foraminal leakage, no leakages to the spinal canal and no cement embolisms. 137

Periprocedural balloon rupture

Neither of the studies of BKP reported periprocedural balloon rupture.

Peri- and postoperative complications (including infection)

Seven studies (Buchbinder et al. ,102 Farrokhi et al. ,135 FREE,137 INVEST,103 Rousing et al. ,128 VERTOS136 and VERTOS II17) provided some information relating to peri- or postoperative complications.

Incidence of new vertebral fractures

Radiographic fractures

Only three studies (Blasco et al. ,127 FREE137 and VERTOS II17) reported the number of patients who suffered new radiographic vertebral fractures during the study period. None of these studies found a statistically significant difference between treatment groups ( Table 14 ). However, in the FREE study,137 loss to follow-up is higher in the control group than in the BKP group [34/149 (23%) vs. 56/151 (37%)]; as the dropout rate outnumbers the event rate in the control group, the fracture incidence data may be biased.

Rousing et al. reported only the number of new radiographic fractures rather than the number of patients who suffered such fractures, and therefore the relative risk of fracture cannot be calculated. They found that, over 12 months, there were more radiographic fractures in the PVP group than in the control group (7 vs. 4, statistical significance not reported). 131

Although the study protocols for the Buchbinder et al. and INVEST studies specified the incidence of new vertebral fractures as an outcome,147,148 the relevant results have not yet been published.

TABLE 14 - Number of patients suffering new incident radiographic vertebral fractures

N/A, not applicable.

Data in normal font were taken directly from the text; data in italics were calculated by the reviewers.

Includes new and worsening fractures.

Mean follow-up: 11.4 months (median 12, range 1–24).

Study	Length of follow-up	Mean time from estimated fracture onset to intervention (weeks)	No. of patients with incident VCF			Relative risk (95% CI)	p-value
			PVP	BKP	Control
Blasco 2012127	12 months	20.0 (13.7)/20.4 (18.6)	17/64 (26.6%)	N/A	8/61 (13.1%)	2.03 (0.94 to 4.35)	0.07
FREE137,138	12 monthsa	5.6 (4.4)/6.4 (5.2)	N/A	38/115 (33%)	24/95 (25%)	1.31 (0.85 to 2.02)	0.22
	24 months		N/A	56/118 (47.5%)	45/102 (44.1%)	1.08 (0.81 to 1.44)	0.62
VERTOS II17	12 monthsb	4.2 (2.4)/3.8 (2.3)	15/91 (16.5%)	N/A	21/85 (24.7%)	0.67 (0.37 to 1.21)	0.18

As noted in Discussion of individual outcome measures included in the review, data from populations with differences in length of follow-up are not directly comparable. However, as all three studies17,127,137 reported results at 12 months, we have performed an exploratory meta-analysis combining data from the three studies which reported the number of patients who had suffered new radiographic vertebral fractures by that time. Although the point estimate favours control, statistical significance was not achieved ( Figure 15 ).

FIGURE 15.

Patients with new incident radiographic vertebral fractures at 12 months. M–H, Mantel–Haenszel.

It has been observed that vertebrae adjacent to those treated with PVP or BKP may be particularly susceptible to subsequent fractures (detailed later in Incidence of new vertebral fractures). Thus, fractures in adjacent vertebrae are more likely to be associated with therapy than fractures in more distant vertebrae. Blasco et al. found that 82% of new fractures in the PVP group were adjacent to the index vertebra compared with 27% in the control group (OR 16.00, 95% CI 1.03 to 835.12, p = 0.0101). 127 The FREE study reported that 28 out of 118 patients in the BKP group (23.7%) and 17 out of 102 in the control group (16.7%) suffered a radiographic fracture adjacent to the index fracture;138 however, the difference was not statistically significant (relative risk 1.42, 95% CI 0.83 to 2.45, p = 0.20). Similarly, Klazen et al. reported that, in VERTOS II, the risk of adjacent rather than distant fracture was not significantly different in the intervention and control groups (p = 0.23), nor did such fractures occur significantly sooner in the PVP group than in the conservative therapy group (4.6 ± 5.4 vs. 6.1 ± 5.9 months, p = 0.48). The only risk factor for either the occurrence or the number of new fractures was the number of vertebral fractures at study entry, which is itself an indicator of the severity of osteoporosis. 186

Other adverse events

The included studies varied considerably in their reporting of other adverse events. Six studies (Blasco et al. ,127 INVEST,103 Liu et al. ,139 Rousing et al. ,131 VERTOS136 and VERTOS II17) did not report any other adverse events. Farrokhi et al. stated only that no emboli occurred;135 these were clearly envisaged as different from cement leakages, which were reported separately.

Buchbinder et al. 102 reported a number of adverse events during the first 6 months of follow-up ( Table 16 ). The figures appear to refer to the number of events, not to the number of patients suffering the event.

The FREE study provided extensive data relating to adverse events. 137,138 Data relating to serious adverse events (defined as adverse events which resulted in death, life-threatening injury or permanent impairment, or which required extended hospital stay or intervention to prevent impairment) are summarised in Table 17 . Few of these serious adverse events were considered to be related to BKP. However, a haematoma which occurred at the surgical site within 2 days of the intervention was considered to be procedure related, as was the exacerbation of a recurrent UTI by catheterisation, also within 2 days of surgery. The patient with the UTI also developed spondylitis near the cement in the vertebral body 376 days after surgery and was treated with antibiotics; however, the inflammation had not resolved by 24 months. None of the adverse events which resulted in death (12 in the BKP group and 11 in the control group) were considered to be related to the device or procedure. 138

Time from fracture to intervention

Of the included studies, only INVEST reported data by baseline pain duration. A post-hoc subgroup analysis of the effect of treatment on pain at 1 month by baseline pain duration categories found no significant difference (p comparing all three categories = 0.58, Table 18 ). 103

As the INVEST study103 was underpowered for this analysis, Staples et al. undertook a meta-analysis of IPD from the INVEST103 and Buchbinder et al. 102 studies to assess the effectiveness of PVP in patients with fracture pain of recent onset (≤ 6 weeks) compared with pain of longer duration. 110 Because the INVEST study103 allowed crossover after 1 month, outcomes were compared only up to that time point. No statistically significant differences in RDQ scores, EQ-5D scores or pain scores were identified between participants whose pain was of recent onset and those whose pain duration exceeded 6 weeks ( Tables 19 to 21 ).

Presence of fracture-related deformity before treatment

No data were identified relating to subgroups with and without fracture-related deformity before treatment.

Receipt of inpatient care before treatment

None of the studies provided information on the number of patients who were inpatients at the time of randomisation, and no data were identified relating to this subgroup.

Baseline pain severity

In the absence of data relating specifically to patients who had received inpatient care immediately preceding the intervention, it may be relevant to note that Staples et al. ’s analyses of IPD from the Buchbinder et al. 102 and INVEST103 studies include patients grouped by baseline pain severity. While p-values were not reported, no statistically significant differences in RDQ scores, EQ-5D scores or pain scores were identified between participants with severe pain (score ≥ 8 on a 0–10 rating scale) or mild to moderate pain (score < 8) at baseline. 110 In both treatment groups, the decrease in pain was greater in the subgroup which had more severe pain at baseline than in the subgroup with less severe baseline pain ( Tables 22 to 24 ), but this presumably simply reflects a greater potential for improvement.

The evidence relating to pain severity prior to PVP therefore suggests that there is no reason to suppose that outcomes would differ between patients who were inpatients prior to treatment and those who were not. This view is strengthened by the fact that receipt of inpatient care following VCF may be influenced by factors other than clinical factors such as pain severity: patients who are bedridden with severe pain may not be hospitalised if they have adequate support networks in terms of both family/friends and community services. No subgroup data are available for BKP.

Registry data

(Academic-in-confidence information has been removed.)

Data from these studies are summarised in Appendix 11 ; they will also be discussed below in the relevant contexts. (Academic-in-confidence information has been removed.) Additionally, a formal critique of the data is provided in Appendix 13 , and summarised later.

Large case series

A total of 10 large case series (n  >  200) provided data on adverse events following PVP or BKP on fractures of osteoporotic origin only. 203–212 One of these studies211 reported data for fractures of non-osteoporotic origin, but the corresponding author provided separate data for osteoporotic fractures, which were included in the review. The adverse event data for BKP and vertebroplasty are summarised in Tables 25 and 26 , respectively.

Serious adverse events related to BKP and vertebroplasty, though relatively rare, are of sufficient importance to warrant consideration in clinical decision-making. Statistical aggregation of data relating to the more common adverse events was not possible because, as shown in Tables 25 and 26 , the data were heterogeneous in terms of what was reported and how it was reported.

All-cause mortality

No deaths were noted in the large observational studies of PVP. However, one procedure-related death was noted by Majd and colleagues205 in a case series of 222 patients who had 360 vertebral bodies treated by BKP. This patient developed an infected shunt and subsequent abscess formation at the site of kyphoplasty. He underwent a discectomy with anterior plus posterior spinal fusion and instrumentation, but did not recover well and subsequently died from cardiovascular failure. It is noteworthy that this patient had previously received a kidney transplant and was taking antirejection medications and prednisone.

(Academic-in-confidence information has been removed.)

A formal critique of the evidence on mortality provided by Medtronic is provided in Appendix 13 , with a summary presented here ( Table 27 ). Observational data can be subject to confounding factors, although methods to adjust for these exist, such as regression analyses using observed variables as covariates and propensity matching. However, neither method can produce a robust estimate of the variable of interest if there is selection of the intervention provided based on unobserved data. Where this may be likely, instrumental variable methods using a variable correlated with an intervention but which is only correlated with the outcome through its effect on the intervention can be employed. However, the validity of an instrumental variable is subjective and can be open to debate.

Evidence on mortality benefit associated with BKP and PVP was submitted by Medtronic in the form of four studies,213–216 all using observational data: two from a claims database from the USA and two from a health insurance fund in Germany. A variety of methods are used, including Cox regression using covariates, matching methods and instrumental variable estimation. The results involved paired comparisons between different groups rather than simultaneous comparisons of the three treatments which may introduce inaccuracy. It is unclear how generalisable these results are to patients treated in England and Wales.

(Academic-in-confidence information has been removed.)

In summary, it is possible that there is a causal difference in mortality between patients treated using OPM and patients receiving BKP or PVP given the size of the effect. Appropriately taking into account the potential endogeneity of the treatment would tend to reduce the point estimate of the effect size but may or may not eliminate it completely. It is not possible to say with certainty if there is a difference in mortality between patients undergoing BKP and PVP as a result of the treatment based on the data presented in the studies included here. There is also considerable uncertainty, were BKP and PVP assumed to have a mortality benefit, regarding whether or not OPLA would also produce a mortality benefit, but no data are available on this.

Note that HRs have been reported rather than statistics such as median or mean survival. This is owing to relatively large numbers of patients remaining alive at the end of the follow-up period. For example, in the Edidin et al. 213 publication median survival had not been reached at 4 years since VCF diagnosis in any of the arms.

Symptomatic and asymptomatic cement leakage

The most common risk associated with vertebral augmentation procedures is cement leakage outside the target vertebral body. The majority of articles reported the incidence of cement leakage in terms of treated vertebral bodies, while four reported it in terms of treated patients. Only Diel’s study of kyphoplasty204 and Lee and Chen’s study of vertebroplasty209 provided leakage incidence data for both treated vertebrae and treated patients. Taken in isolation, data relating to either the number of vertebrae or the number of patients are potentially misleading and could introduce systematic bias towards under-reporting of incidence.

The location of cement leakages has important implications for safety; intradiscal leakages are unlikely to lead to morbidity, but leakages into the epidural space or venous system have the potential to cause serious complications. 217 Three studies (Majd,205 Diel207 and Evans208) did not report the location of cement leakages. Poor reporting of follow-up duration and completeness was also a problem for interpreting these data. Lee and Chen209 did not report follow-up duration, while in most of the other studies it was unclear what proportion of the cohort was lost to follow-up at what time points, and why.

When reported in terms of treated vertebral bodies, the incidence of leakage ranged from 5%210 to 72%206 for vertebroplasty, and from 9%204 to 18%205 for kyphoplasty. By contrast, when reported in terms of treated patients, leakage incidence was higher, ranging from 12%211 to 87%212 for vertebroplasty. Only one kyphoplasty study204 reported incidence in terms of treated patients: it reported a rate of 19%. It is not clear why such wide variations in incidence were observed, but factors such as practitioner skills and experience, clinical setting, cement viscosity and thoroughness of follow-up may have played a part.

Epidural leaks appeared to be common in vertebroplasty cohorts. Ryu and Park212 reported epidural leaks in 157 of 215 treated patients (73%), and Álvarez et al. 206 reported three leaks into the spinal canal and 132 into the epidural veins in a cohort of 260 patients with 423 treated vertebrae (52% of patients). These complications did not appear to be as common in kyphoplasty cohorts. Blattert et al. 203 and Diel et al. 204 reported six and four leaks into the epidural space in cohorts of 314 and 320 respectively (2% and 1% respectively). Because of the nature of follow-up in the cohorts, the long-term clinical implications of these cement leaks are unknown. Several investigators undertook long-term follow-up: Masala and colleagues210 reported a follow-up duration of up to 3 years, although only 68 patients (24%) had data available at that time point; Majd et al. 205 reported a follow-up duration of up to 36 months, although data were not available on how many patients had data available at given time points; Blattert and Josten203 reported a minimum 2-year follow-up and did not report any missing data; and Álvarez et al. 206 reported follow-up of up to 96 months though, again, it was unclear how many patients were followed up at particular time points.

Other reported adverse events which may be related to cement leakage included pulmonary embolism,207,218 radiculopathy204,205 (which in the study by Diel et al. 204 was specifically due to cement extrusion), temporary radicular pain208 and temporary and permanent motor deficits or parapesia of the legs. 206

Intraoperative balloon rupture

Intraoperative balloon rupture appears to be a relatively rare complication of BKP: in the two studies which report it (Blattert et al. 203 and Diel et al. 204), it occurred in 6 out of 352 procedures (1.7%) in Blattert et al. ’s cohort,203 and in 1 out of 391 (0.3%) in Diel et al. ’s study. 204 Neither of these studies discussed the clinical implications of balloon rupture. However, Saliou et al. 219 discussed some of the potential implications in a smaller case series in which it was more common (n = 51, treated levels: 75, balloon rupture in 5 vertebrae of 5 patients). Although no symptomatic complications due to balloon rupture were observed in that study, the authors point out that this complication could lead to contrast leakage, procedural delay or gas embolism.

Other peri- and postoperative complications (including infection)

In the included case series, peri- and postoperative complications were relatively rare. Majd et al. 205 reported 10 medical and three surgical complications in 222 patients undergoing BKP. Most of the medical complications related to pre-existing cardiac, pulmonary or liver disease. In one case, a patient was treated with local anaesthesia because medical comorbidities made general anaesthetic inadvisable, and developed electrocardiogram abnormalities during the procedure; treatment of a second VCF had to be postponed for 4 days while the patient was assessed by a cardiologist. In addition to one case of infection discussed under All-cause mortality, above, and one cement leak causing radiculopathy, also discussed above, the surgical complications included one patient who needed surgical debridement, irrigation and closure of the wound 3 weeks after BKP. 205

Diel et al. reported one instance where the vertebral wall was fractured, with displacement of the balloon catheter, in a patient undergoing BKP,204 and eight cases of temporary hypotension following cement injection in 202 patients undergoing PVP for osteoporotic fracture (3.9%). 207

Incidence of new vertebral fractures

New vertebral fractures have been identified as an important source of postoperative morbidity among people with osteoporotic VCFs treated with PVP or BKP. The observational studies by Harrop et al. ,220 Kulcsar et al. ,221 Tseng et al. 222 and Uppin et al. 223 specifically set out to study the overall incidence of new vertebral fractures in osteoporotic patients following PVP or BKP. However, as these patients are by definition at increased risk of vertebral fracture, the data are difficult to interpret in the absence of a control group of similar patients who have not undergone PVP or BKP. Similarly, although a number of retrospective reviews of new vertebral fractures in patients treated with PVP or BKP were identified185,220–236 and reported incidence rates ranging between 6.8% over a 25.6-month follow-up period229 to 22.2% during a 1-year follow-up,233 it is difficult to know how to interpret these data. However, it should be noted that, because new fractures were generally identified only when patients returned to clinic with recurring back pain, the reported figures probably represent a conservative estimate of true fracture incidence.

Arguably of greater relevance was the finding from the case series by Harrop et al. ,220 Kulcsar et al. ,221 Tseng et al. 222 and Uppin et al. 223 that new VCFs are significantly more likely to occur in vertebrae adjacent to treated levels than in non-adjacent vertebrae. Although two reports found a similar crude incidence rate of adjacent and non-adjacent fractures,232,234 it should be added that patients would typically have a greater number of non-adjacent vertebral bodies that could fracture, so even these data may represent a greater likelihood of fracture at adjacent levels. 236 In addition, some studies show that, following vertebral augmentation, adjacent fractures are likely to occur sooner than non-adjacent fractures. Donovan et al. 224 reported the case of a 50-year-old woman who developed several new fractures 8 days after a kyphoplasty procedure, and concluded that the ‘temporal relationship between the kyphoplasty procedure . . . with documented fractures of six adjacent vertebrae . . . is highly suggestive of causality’ (p. 712). A larger retrospective analysis of time between vertebroplasty and new adjacent fractures234 found times to diagnosis of new adjacent and nonadjacent fractures of 55 and 127 days, respectively (p < 0.0001). Further evidence was supplied by Mudano and colleagues,237 who compared the rate of new fractures in a cohort of patients treated with PVP or BKP against a cohort of patients with VCFs and no cement augmentation. A significantly higher incidence was observed in the treated cohort at 90 days (adjusted OR 6.8, 95% CI 1.7 to 26.9) and 360 days (adjusted OR 2.9, 95% CI 1.1 to 7.9).

A number of prognostic factors have been associated with higher risk of subsequent vertebral fractures: these include increased age and number of treated vertebrae,228 presence of clefts in the treated VCFs,235,238 and spinal instability measures. 227 There is also a growing body of evidence suggesting biomechanical explanations for the higher rate of adjacent fractures. A number of studies have demonstrated that bone cement can increase the stiffness of the treated vertebra, resulting in an increase in loading on the adjacent vertebrae. 239–242 Cement leakage may also play a part: Han et al. 225 found that, when adjacent VCFs occurred, fractures were more likely to be close to extraneous cement. In contrast to these studies, Farooq et al. 243 demonstrated that vertebroplasty could partially reverse fracture-induced changes including decompression of the adjacent nucleus and higher neural arch load-bearing.

However, in the absence of well-controlled randomised studies, neither time from surgery to new VCF nor a higher incidence of adjacent vertebral fractures (compared with non-adjacent fractures) can be considered as definitive evidence of causation.

Rib fractures

Two large observational studies reported rib fractures related to vertebroplasty: Álvarez et al. 206 reported five fractures in a cohort of 260 patients, while Evans et al. 208 reported seven in a cohort of 245. No rib fractures were reported in the kyphoplasty case series.

Refracture of treated vertebrae

It has been suggested that a treated vertebra may refracture either because too little cement was injected or because the vertebra was extremely fragile and therefore at risk of refracture even when adequate quantities of cement were injected. 187 However, as none of the included case series reported this complication, incidence is likely to be low.

Percutaneous vertebroplasty and BKP may be associated with transitory increase in post-procedural pain. However, among the large case series, only Evans et al. 208 reported this complication: three patients from a cohort of 245 experienced worsening of pain, although no biomechanical causes could be found. While the other case series did not report worsening of pain as an adverse event, it was unclear whether this was because it did not occur or because the authors did not view transitory increases in pain as an adverse event per se.

Case reports

In general, case reports were included only if they reported adverse events which had not been reported in the larger observational studies. However, an exception was made in the case of pulmonary cement embolism, the most commonly reported complication of vertebral augmentation, where all identified case reports were included in order to indicate subsequent therapy, if any.

Pulmonary cement embolism

The search identified 46 case reports245–290 and 47 patients in whom a pulmonary cement embolism caused by venous PMMA leakage was detected. These reports related to 41 vertebroplasty and five kyphoplasty procedures. Four deaths due to pulmonary embolism were reported in vertebroplasty patients; no deaths were identified in kyphoplasty patients. Sixteen embolisms were reported as asymptomatic, while 29 were symptomatic; for the remaining two, no details were provided on symptomatology. Symptomatic manifestations of pulmonary embolism include dyspnoea, tachycardia, chest pain, dizziness and sweating. Asymptomatic pulmonary embolism is more difficult to detect and, furthermore, it is difficult to gain understanding the long-term clinical implications of these silent pulmonary emboli from the available data. The case study data relating to pulmonary embolism are summarised in Table 29 .

Postoperative infection

A number of case reports291–302 have described postoperative infectious complications. While such infections can occasionally be managed with a medical approach, they often necessitate further surgical intervention. 293,294,296–298,300,302 One team292 reported the death of a patient from septic multiple organ failure after antibiotic treatment and local surgical interventions.

Other adverse events

A number of case reports noted rare but serious cardiovascular complications related to vertebral cement augmentation, including cardiac perforation,281,285,303,304 inferior vena cava syndrome,305 venous air embolism,306 vena cava thrombus,307 acute pericarditis,308 lumbar artery pseudoaneurism309 and stroke. 310

Biafora et al. 311 reported an injury to a segmental branch of the L4 lumbar artery in an 84-year-old patient: this manifested clinically in bleeding from the kyphoplasty site and was successfully treated with torpedo embolisation of a small branch of the right L4 lumbar artery. Heo and Cho312 reported a L2 segmental artery injury, which was also successfully treated with endovascular embolisation. Hard et al. 313 reported a transpedicular needle penetrating the margin of the T5 vertebral body by 15 mm. Injury to the posterior aortic wall was confirmed and an improvised injection of PMMA was used to seal the aortic wall. No related complications were seen during 2 years of follow-up.

Ozturk et al. 314 reported a case of irreversible complete paraplegia due to cement leakage into the spinal canal. Lee et al. 315 presented a case of complete motor and sensory deficits at T11 due to cement leakage, which was treated with surgical decompression. Birkenmaier et al. 316 reported a transitory paraplegia in an 82-year-old patient following a massive epidural haematoma compressing the cauda equine and the conus medullaris. The haematoma was drained, resulting in the loss of 3 litres of blood and requiring transfusion of packed red blood cells and fresh-frozen plasma. However, 48 hours post procedure, full neurological function had been regained. Lopes and Lopes317 also reported paraplegia due to spinal cord and root compression, which was successfully remedied with surgical decompression.

Lim et al. 318 reported two cases of subarachnoid haemorrhage: both patients were treated successfully with medical management. Other rare complications included heterotopic ossification,319 addisonian crisis,320 lumbar disc herniation,321 posterior spinal epidural abscess322 and a fatal fat embolisation with no evidence of cement leakage. 323

Summary and discussion of data relating to adverse events

The evidence drawn from the included RCTs, case series, and case reports suggests that PVP and BKP may be associated with a number of adverse events. Treatment-related deaths appear to be rare, but cement leakage is common, particularly with PVP: pooled data from the RCTs indicate an incidence of 44% of treated vertebrae for PVP and 27% for BKP (see Table 13 ), while the case series indicate a range of 5% to 72% for PVP and 9% to 18% for BKP (see Tables 25 and 26 ). While many cement leaks were not associated with immediate clinical complications, others were associated with serious problems such as pulmonary embolism, radiculopathy, and temporary or permanent motor deficits. A number of procedure-related deaths have been noted. Moreover, there is as yet no good evidence to prove that that leaks which are asymptomatic in the short term do not have long-term implications.

Peri- and postoperative complications other than cement leak appear to be rare, though potentially serious. In particular, infectious complications are potentially fatal and frequently require treatment with further surgical intervention. To reduce the risk of such complications, it has been recommended that PVP or BKP should not proceed until the patient has made a complete recovery from any existing infections, and that, in cases of recent infection, either antibiotics should be prescribed on a long-term basis to avoid deep infection or a cement–antibiotic mixture should be used. 293 Intraoperative balloon perforation during kyphoplasty seems unlikely to lead to any serious complications. Nevertheless, Saliou et al. 219 have suggested a number of methods to minimise the incidence of rupture: (1) purge any trapped air from the balloon to prevent gas embolism, (2) increase balloon inflation pressure very slowly, to allow the balloon to adapt to the solid, sharp, bony environment, and (3) use a curette to break bone bridges and constitutive bone fragments before inflating the balloon.

While it seems likely that PVP and BKP may be associated with increased rates of new vertebral fractures, and in particular adjacent fractures, as yet the quality of the evidence for this is not good.

It is also unclear which of PVP or BKP is the safer of the two approaches to vertebral augmentations, as direct comparisons were unavailable. However, Yang et al. 324 conducted a review which found that rates of specific complications (cement leakage, new compression fractures, pulmonary embolism and radiculopathy) were all significantly higher with vertebroplasty than with kyphoplasty (all p < 0.05). They also found that cement leakage rates were lower in procedures carried out in neurosurgery departments (20.6%) and orthopaedic departments (24.7%) than in radiology departments (52.9%). This could, however, be confounding if the vertebroplasties were carried out by the radiologists. In addition, Medtronic claim that, in BKP, the creation of a cavity within the vertebral body allows for the insertion of a pre-known volume of a more viscous cement at a lower pressure, which reduces the risk of cement leakage and consequent complications compared with PVP. 34

None of the included studies referred to the radiation risks to patients associated with PVP and BKP. These risks, though low, are not trivial. Perisinakis et al. 325 estimated, on the basis of a case series of 11 patients undergoing kyphoplasty with fluoroscopic guidance, a rate of 741 fatal cancers and 5.4 hereditary effects per million treated patients. However, Fitousi et al. 326 found a relatively high level of radiation exposure in a case series of 11 patients undergoing vertebroplasty with fluoroscopy, and estimated a fatal cancer risk of 1 in 580 and a risk of hereditary effects of 1 in 20,000.

Finally, it should be noted that, although PVP and BKP may be associated with the adverse events discussed above, the alternative treatment (conservative management with analgesics, back bracing and bed rest) is linked to a number of potentially serious complications. Bed rest can lead to muscle wasting and deconditioning, and these effects have been associated with DVT, pulmonary emboli, reduced muscle blood flow, red cell volume, capillarisation and oxidative enzymes. 327,328 Narcotic analgesics are associated with a number of undesirable side effects including cognitive impairment and nausea, while NSAIDs are associated with gastrointestinal problems. 55 The registry studies indicate that (academic-in-confidence information has been removed) (for details, see Appendix 11 ). (Academic-in-confidence information has been removed.)

Internal validity

The evidence for the clinical effectiveness of PVP is not consistent. The best quality studies, the blinded Buchbinder et al. 102 and INVEST103 studies, show no benefit, whereas some benefit is seen in the lower-quality unblinded studies.

The unblinded FREE137 study suggests some benefit from BKP, although any benefits diminished over time.

Various suggestions have been put forward to explain the inconsistency between the results of the blinded and unblinded studies of PVP. These suggestions relate to:

• patient selection (fracture acuity and pain severity)

• operator technique (volume of injected cement and/or technique used for injection)

• nature of the OPLA

• outcome measurement

• use of blinding.

These are discussed in turn below.

Patient selection

Operator technique

The technique used in the Buchbinder study to inject cement (i.e. 13-gauge needles and cement hand-injected using 1-cc syringes102,148) has been criticised on the grounds that, to achieve adequate filling of the vertebral body in lumbar fractures, either an 11-gauge needle or a high-pressure injecting system should have been used. 113

It has also been suggested that the volume of cement injected by Buchbinder et al. was too low. 21,113 Aebi has also criticised the INVEST study103 on this basis, suggesting that as the mean amount of cement injected in both studies was inadequate, the investigators in essence compared two placebo operations. 21 However, Kaufmann et al. 330 noted that greater cement volumes may have better outcomes but higher risks of adverse events. Moreover, although Al-Ali et al. 331 found a mean volume of cement injected of 5.1 ± 2.2 ml in 600 osteoporotic fractures treated by vertebroplasty, the range was 1.0 ml to 16.0 ml, and no correlation was observed between the volume of cement and pain improvement. However, they noted that the volume injected was sufficient to fill the intravertebral cleft and, as long as that was done, the volume of cement used was not a determining factor in the degree of pain relief. 331

External validity

Several factors may affect the external validity of the included studies. The first relates to the potential learning curve relating to the vertebral augmentation procedures. It seems reasonable to assume, and indeed is in some cases stated, that the procedures reported in those studies were performed by experienced personnel, and therefore their results may differ from those obtained by less experienced practitioners. There is little evidence to indicate how many procedures a practitioner needs to perform to achieve a high standard. McDonald et al. compared the outcomes of PVP performed in the Mayo Clinic, MI, USA, by two interventional neuroradiologists with substantial previous vertebroplasty experience and five experienced interventional neuroradiologists who were initially new to the procedure; the ‘experienced’ operators estimated that they had performed at least 150 PVPs prior to the commencement of the study. Patient outcomes appeared to be broadly similar regardless of operator experience, although loss to follow-up limited exploration of long-term outcomes. However, both the volume of cement used and postoperative pain (as assessed by pain at rest and RDQ scores 1 week after PVP) were higher with ‘novice’ than with experienced operators, but decreased as the ‘novice’ operators gained more experience. 341 Thus, although a learning curve can be observed, its effects seem to be relatively limited. However, as the authors caution, the ‘novice’ operators in their study were all highly skilled interventional radiologists with substantial clinical experience prior to the study, and the results are therefore not necessarily generalisable to less skilled personnel. 341

There is potentially an issue about the type of cement used in the studies, as newer generation cements (including those which are highly viscous) have been designed to reduce the risk of cement leakage. 36

• Syed et al. conducted a RCT of PVP in patients with osteoporotic VCF comparing PMMA with Cortoss,™ a bioactive composite – but this abstract342 does not report that comparison.

• Blattert et al. 343 RCT compared BKP in patients with osteoporotic VCF (including burst fractures) with PMMA and Norian SRS, a calcium phosphate/carbonate cement. Each cement was associated with leaks in 5 out of 30 vertebrae. PMMA was associated with vascular embolism in two patients; however, with Norian SRS, there were nine cases of cement failure, (i.e. at follow-up at 6 weeks they showed radiographic signs of early cement fracture) all in burst fractures. The investigators therefore did not recommend its use in BKP (there was also persistent haemorrhaging from one vertebral body which partially washed out the cement before it could set – this is less likely to happen with PMMA). (An advantage of calcium phosphate-based cements is that they set at a significantly lower temperature than PMMA, and therefore there is less risk of thermal damage to adjacent structures.)

• Anselmetti et al. 344 in a single-centre RCT compared PVP with standard low-viscosity PMMA and high-viscosity PMMA designed for injection through a proprietary delivery system (Confidence Type I, Disc-O-Tech, Israel) in patients with VCF of any origin; all procedures were performed by the same experienced operator. CT scans were performed 1 hour after PVP to evaluate cement perfusion, leakages and possible complications; when a venous leak was detected, a CT scan of the lungs was performed to assess the possibility of PMMA embolism. No symptomatic cement leaks occurred in either group; asymptomatic venous leaks were significantly less common in vertebrae treated with high-viscosity PMMA than in those treated with low-viscosity PMMA, but the reduction in the number of leaks into the disk was not statistically significant ( Tables 30 and 31 ).

Summary

The included studies measured back-specific functional status using a number of instruments, including the RDQ and SOF-ADL, and indicators of disability such as walking aids. The FREE study137 reported significantly better RDQ outcomes in the BKP group at 1 month and 12 months, although the 95% CIs included the possibility of a lack of MCID at both time points. The VERTOS II study17 reported a significant difference favouring PVP at 12 months. However, the two blinded OPLA-controlled studies102,103 found no statistically significant between-group differences. Farrokhi et al. measured functional status with an instrument based on the ODI, and found a statistically significant difference favouring PVP at all follow-up time points. 135

Five studies (Buchbinder,102 FREE,137 INVEST,103 Rousing131 and VERTOS II17) measured quality of life using the EQ-5D. However, VERTOS II did not report follow-up values, and both Rousing and Buchbinder only began to collect EQ-5D data part way through the trials. Aggregation of IPD from the Buchbinder and INVEST trials110 found no significant difference in EQ-5D in short- or medium-term outcomes. Conversely, the FREE trial found significant differences favouring BKP throughout follow-up. Another commonly used quality of life measure was QUALEFFO, which was reported in four studies (Blasco et al. ,127 Buchbinder et al. ,102 VERTOS136 and VERTOS II17). Broadly speaking, there was a tendency towards favouring PVP. However, the OPLA trial found no significant between group differences on any of the QUALEFFO subscales at any time point.

Four open-label studies (Farrokhi et al. ,135 FREE,137 Rousing et al. 131 and VERTOS II17) found statistically significant differences between groups in short- and medium-term improvement in pain. Conversely, the two double-blinded, OPLA-controlled studies of vertebroplasty, Buchbinder et al. 102 and INVEST103 found no statistically significant between-group differences in pain, and the trend towards a higher rate of clinically meaningful improvement in the INVEST trial may be confounded by greater opioid use in the PVP group. INVEST103 also measured pain frequency and pain bothersomeness on a scale of 0 to 4 and did not find a statistically significant, or clinically meaningful, difference between groups. A similar picture emerged for analgesic use. INVEST and Buchbinder reported reductions in opioid use from baseline in both the PVP and control group, with no statistically significant between-group differences at any time point. However, VERTOS and VERTOS II reported greater short-term reductions in analgesic use among patients treated with PVP, while short- and medium-term differences favouring BKP were reported in the FREE trial.

None of the studies reported statistically significant differences in mortality between treatment groups. However, none of the studies were powered to detect this outcome. A meta-analysis was performed on the three studies which reported all-cause mortality at 12 months. 17,127,131 Although the pooled result slightly favoured PVP, the effect failed to reach statistical significance. Furthermore, there are plausible biomechanical explanations as to why vertebral augmentation may increase life expectancy; these include improvement of lung function due to correction of kyphotic deformity345 and mitigation of impaired physical function through pain relief. 345,346 However, the absence of randomisation in this cohort means that confounding factors cannot be ruled out. A formal critique of mortality data from observational databases (see Appendix 13 ) concludes that:

It is possible that there is a causal difference in mortality between patients treated using OPM and OP patients given the size of the effect. Appropriately taking into account the potential endogeneity of the treatment would tend to reduce the point estimate of the effect size but may or may not eliminate it completely. It is not possible to say with certainty if there is a difference in mortality between patients undergoing BKP and PVP due to the treatment based on the data presented in the studies included here.

There is also considerable uncertainty, were BKP and PVP assumed to have a mortality benefit, in whether or not OPLA would also produce a mortality benefit. However, there were no data on this.

Percutaneous vertebroplasty and BKP appear to be reasonably safe procedures, with a low rate of intra- and postoperative complications. However, when complications do arise from vertebral augmentation, they can be serious. Among the several large case series analysed in this review, one reported a death, though it seemed likely that factors other than the vertebral augmentation played a part. 205 However, several deaths directly related to augmentation procedures have been noted in case reports. Cement extrusion was common, including extrusion into the epidural space. However, most reported extrusions remained asymptomatic. Vertebral augmentation does seem to be associated with a higher risk of new adjacent fractures. A substantial number of case reports of pulmonary embolism, and several case reports of rare but potentially serious complications were identified.

Discussion of potential subgroups

The included studies were too small to permit the identification of subgroups of patients who might benefit from PVP or BKP. While there has been considerable debate on whether vertebral augmentation is more effective in the treatment of acute or chronic fractures,334,336,347,348 analysis of individual patient data from the two double-blinded RCTs of PVP suggested that effectiveness was unrelated to fracture acuity. 110

It has been suggested that PVP and BKP are more successful in patients with a mobile pseudarthrotic cleft pattern of fracture than in those with the more common non-mobile fracture,202 but further research is required to explore this possibility. It has also been suggested that there is a substantial subgroup of patients whose pain is not a direct result of VCF but of overload of facet joints, paraspinal muscles and impingement of spinous processes; such patients respond to facet joint injection while those who do not respond to this treatment have an excellent response to vertebroplasty. 84 However, further evidence from clinical trials would be required to confirm the importance of such subgroups.

The assumed acquisition costs

The list prices for PVP using the CONFIDENCE SPINAL CEMENT SYSTEM™ were taken from the Johnson & Johnson submission. 36 The costs vary according to the number of levels that need to be treated and are reported to be £1358 at one level, £1784 at two levels and £1848 for a three-level approach. It is noted that table 83 of the Johnson & Johnson submission appears to contradict the text regarding the cement required in the two-level procedure. 36 The text states that 11 cc would be required, whereas table 83 assumes cc were sufficient. The distribution of operations between one, two and three levels were extracted for Johnson & Johnson by Dr Foster, and are 58.9%, 20.5% and 20.5%, respectively. The costs per level were multiplied by these proportions to arrive at a weighted cost of £1472. If 11 cc of cement were assumed in the two-level operation rather than 7 cc this value would increase to £1546.

The Johnson & Johnson submission36 inflated the price of a BKP operation reported in Ström et al. 31 and assumed a cost of £2842 for BKP. The manufacturer also assumed that the cost of OPLA would equal the cost of PVP; this may be questionable, particularly when it is assumed that high-viscosity cement would be used in the OPLA rather than cheaper low-viscosity cement.

The costs of the preliminary phase, the operating phase and the postoperative phase have previously been reported in Ström et al. 31 These were inflated to 2009–10 prices within the Johnson & Johnson submission,36 and are partially replicated in Table 35 . The table in the Johnson & Johnson appears to misreport the operating room costs, which should be £275, as used in their mathematical model. This has been amended in Table 35 . It was assumed that these costs are applicable to both BKP and PVP.

The assessment group comments that there appeared to be a typographical error in the manufacturer’s mathmatical model in which only 10% of patients receiving BKP were assumed to consume operating room resources; it was assumed that this value was intended to be 100%. As such the overall cost-effectiveness results are likely to be favourable to BKP.

Johnson & Johnson also undertook bottom-up costing within the operating phase. This approach used data on procedure times and number of vertebral levels treated, collected using a bespoke iPad (Apple Inc., Cupertino, CA, USA) application designed to measure the total duration of the operating room episode in minutes. These data were collected at five hospitals (details confidential). Data from the iPad app were used in conjunction with data from an audit of vertebroplasty procedures, obtained from two hospitals (identity confidential) currently offering vertebroplasty, to generate estimated costs for the procedure. While the average weighted operating cost per procedure can be made public, the breakdown of constituent parts remain confidential. Data obtained included the cost and volumes of surgical consumables, medication costs (including sedation and antibiotics), theatre costs and staff costs. The data from this analysis are replicated in Tables 36 and 37 .

The costs associated with hospitalisation stay

Results presented by Johnson & Johnson

The base-case deterministic results are replicated in Table 39 . It is seen that PVP was shown to dominate (that is to say, producing more QALYs at a lower cost) BKP. The cost-effectiveness plane and cost-effectiveness acceptability frontier are shown in Figure 16 . The incremental cost-effectiveness ratio (ICER) between PVP and OPM was £4392 per QALY gained. Explicit comparison between the PVP and BKP results indicated that the ICER between PVP and BKP was 99.86% likely to be below £20,000 per QALY gained.

FIGURE 16.

The cost-effectiveness plane and frontier associated with the base-case deterministic results in the Johnson & Johnson submission. 36 CE, cost-effectiveness.

The cost-effectiveness frontiers and ICERs for the target population and the alternative scenarios undertaken by Johnson & Johnson are summarised in Table 40 .

Univariate sensitivity analyses

Johnson & Johnson undertook univariate sensitivity analysis comparing PVP with OPM and PVP with BKP. 36 The analyses for PVP and OPM are reproduced in Figure 17 . The analyses comparing PVP and BKP are not reproduced as in all analyses undertaken PVP dominated BKP.

FIGURE 17.

A tornado plot of univariate sensitivity comparing PVP with NIM in the base case. NIM, non-invasive management; VP, vertebroplasty.

A threshold analysis was undertaken on the parameters contained in Figure 17 in the base case. These values are reproduced in Table 41 . Similar analyses were presented for the target population with broadly similar results.

TABLE 41 - Threshold analysis on key parameters affecting the deterministic results in the Johnson & Johnson submission36

KP, kyphoplasty; NIM, non-invasive management; VP, vertebroplasty.

Indicates an illogical parameter value.

Variable	Base case	£20,000 per QALY	£30,000 per QALY
Ström:31 cost of hospital stay (days)	£232	–£63a	–£251a
Costs of VP procedure	£1479	£4239	£6008
Ström:31 hospital stay (days) (VP)	3.24	15.14	22.76
Ström:31 hospital stay (days) (OPM)	12.61	0.71	–6.91
EQ-5D regression: VAS coefficient	–0.10	–0.02	–0.01
VAS at 1 month for VP	2.53	8.79	9.38
VAS at 6 months for VP	2.41	5.56	5.86
VAS at 1 month for OPM	4.94	–1.32a	–1.91a
VAS at 12 months for VP	2.17	7.95	8.49
VAS at 6 months for OPM	4.11	0.96	0.66

The assumed acquisition costs

The list price for a BKP kit (£2600.50) has been taken from the Medtronic submission;34 Medtronic additionally quote a lower price as an average selling price but this value (£1900) is not consistently available to all customers within the NHS. An additional £96 has been added for devices used in the operation.

Medtronic assume a cost of PVP of (commercial-in-confidence information has been removed) reported to be the average selling price of De Puy’s spine PVP plus an additional £53 for devices used in the operation.

The costs associated with the operation

The costs of the preliminary phase, the operating phase and the postoperative phase have previously been reported in Ström et al. 31 Medtronic updated these costs in table 49 of their submission, which are replicated in Table 43 .

The costs associated with hospitalisation stay

The results presented by Medtronic

Probabilistic results

The probabilistic results are replicated in Table 49 . When compared with the results in Table 45 it is seen that deterministic and probabilistic results were similar. The assessment group notes that it is unclear why results for total life-years gained were not reported in the probabilistic analyses.

TABLE 49 - Probabilistic base-case results presented in the Medtronic submission34

LYG, life-year gained; NSM, non-surgical management.

Technologies	Total costs (£)	Total LYG	Total QALYs	Incremental costs (£)	Incremental LYG	Incremental QALYs	ICER (£) vs. NSM (QALYs)	ICER (£) incremental (QALYs)
Probabilistic analysis
OPM	5394		4.975
PVP	6132		5.327	738	0.00	0.35	2100	2100
BKP	6385		5.443	991	0.00	0.47	2118	2174

A cost-effectiveness acceptability curve (CEAC) was submitted by Medtronic, which is reproduced in Figure 19 .

FIGURE 19.

The CEAC presented in the Medtronic submission. 34 Reproduced with permission from Medtronic.

A further comparison of the results between BKP and PVP was provided (and replicated in Figure 20 ). This indicated that in the large majority of cases BKP provided more QALYs than PVP, with an approximately even distribution between whether BKP or PVP was the more expensive procedure.

FIGURE 20.

A scatterplot of the paired BKP and PVP results. Reproduced with permission from Medtronic.

To ‘patient sustains an additional vertebral fracture and remains alive’

The transition rates were taken from the values used in Stevenson et al. 357 These values were derived from an exponential fit to population data provided in Singer et al. 37 and adjusted to provide a risk for patients with a T-score of –2.5 SD and no previous fracture (table 20 of Stevenson et al. ). These values were then multiplied by 1.5 to take into account the additional risks following an initial fracture for patients aged 70 years or over. Further detail on this calibration is provided on p. 43 and illustrated in figure 23 of Stevenson et al. 357

Alternative T-scores were considered by using the equations provided by Marshall et al. ,54 which indicate that risk of vertebral fracture increased by 1.8 to the power of the patient’s Z-score. The Z-score is defined as the number of SDs from the average BMD of women of the same age as the patient. Assuming that the SD of BMD remains constant as a population ages, the risk of fracture at a T-score of –3.5 SD was therefore assumed to be 1.8 times greater than that of a patient with a T-score of –2.5 SD.

The annual vertebral fracture risks assumed for patients of a given age, and given T-score on entry to the model, are provided in Table 50 .

To take into consideration that a patient’s bone density is likely to deteriorate over time, a decrease of 0.255 SD per 5-year age group was incorporated in accordance with data from Holt and Khaw338 presented in Stevenson et al. 359 Thus, when a patient became 5 years older the risk of a vertebral fracture increased by 1.8^0.255, which is an increase of 16% compared with a person of the same age, with a T-score equal to that of the patient 5 years previously. For simplicity, it was assumed that women and men with the same T-score would have the same risks of fracture.

If a patient were assumed to be taking a bisphosphonate, the effect on vertebral fractures was assumed to be a relative risk of 0.58 (95% CI 0.50 to 0.67) from data reported in table 27 of Stevenson et al. 357 This effect was assumed to last for 5 years, with a linear wane in effect over a 5-year period, so that the relative risk was 1 after 10 years.

To ‘patient sustains a hip fracture and remains alive’

The transition rates were taken from the values used in Stevenson et al. 357 These values were derived from an exponential fit to population data provided in Singer et al. 37 and adjusted to provide a risk for patients with a T-score of –2.5 SD and no previous fracture (table 20 of Stevenson et al. 357). These values were then multiplied by 1.5 to take into account the additional risks following an initial fracture for patients aged 70 years or over. Further detail on this calibration is provided on p. 43 and illustrated in figure 23 of Stevenson et al. 357

Alternative T-scores were considered by using the equations provided by Marshall et al. ,54 which indicate that risk of vertebral fracture increased by 2.6 to the power of the patient’s Z-score. Assuming that the SD of BMD remains constant as a population ages, the risk of fracture at a T-score of –3.5 SD was therefore assumed to be 2.6 times greater than that of a patient with a T-score of –2.5 SD.

The annual risks of hip fracture assumed for patients of a given age, and given T-Score on entry to the model, are provided in Table 51 .

To take into consideration that a patient’s bone density is likely to deteriorate over time, a decrease of 0.255 SD per 5-year age group was incorporated in accordance with data from Holt and Khaw358 presented in Stevenson et al. 359 Thus, when a patient became 5 years older the risk of a vertebral fracture increased by 2.6^0.255, which is an increase of 28% compared with a person of the same age, with a T-score equal to that of the patient 5 years previously. For simplicity, it was assumed that women and men with the same T-score would have the same risks of fracture.

If a patient were taking a bisphosphonate, the effect on hip fractures was assumed to be a relative risk of 0.72 (95% CI 0.58 to 0.88) from data reported in table 27 of Stevenson et al. 357 This effect was assumed to last for 5 years, with a linear wane in effect over a 5-year period, so that the relative risk was 1 after 10 years.

To ‘death through fracture or non-fracture related causes’

The mortality rate associated with hip fracture was taken from table 21 of Stevenson et al. ,357 who report that an estimated 6% of people aged 70–79 years living in the community die from causes related to a hip fracture in the year of fracture. Corresponding figures for patients aged 80–89 years and 90 years or over were 11% and 16%, respectively. For simplicity, it was assumed that all patients lived in the community prior to the osteoporotic VCF.

The mortality rate associated with vertebral fracture was taken from a UK study360 comparing mortality in those with osteoporosis (and no fracture) with mortality in those with osteoporosis and a previous clinically apparent vertebral fracture. The HR was 4.4 (95% CI 1.85 to 10.6) and was used in the model to inflate the underlying death rate through other causes. The number of years for which a vertebral fracture was assumed to affect mortality rates was user defined with a base-case estimate of 5 years. The effect then linearly dissipated across a user-defined period (5 years in the base case). When a patient was simulated to have a subsequent vertebral fracture, the model had the facility to allow an increase risk of mortality in the year of subsequent fracture in accordance with data from Jalava. 360 Any effects in subsequent years were not incorporated to limit the number of health states required.

It was assumed that the mortality rate following hip fracture could not be lower than either the mortality rate associated with a vertebral fracture or that of general mortality in the underlying age- and sex-matched population. In such circumstances the rate of mortality following hip fracture was increased to equal the higher value.

The underlying death rate through other causes than fracture was taken from the Office of National Statistics’ Interim Life Tables. 361 For simplicity, it was assumed that all patients would die in their 101st year.

There has been published evidence that mortality may be influenced by initial procedure213 and further data have been provided in the submission by Medtronic. 34 This is critiqued in Appendix 13 . Where BKP, PVP or OPLA were assumed to have positive mortality effects compared with non-invasive management (NIM), these were incorporated in the model for a user-defined period (set to 5 years in the base case). It was assumed that mortality benefit does not wane in a linear fashion, but would cease immediately after the user-defined period. The relative risks associated with treatment were assumed to apply to the all-cause mortality rates and to the increase associated with vertebral fractures, but not to the value following hip fracture.

To ‘patient sustains an additional vertebral fracture and a hip fracture and remains alive’

It was assumed that the risk of hip fracture for patients was independent of whether or not the patient was simulated to have a subsequent vertebral fracture. Therefore, the methodology for calculating the risk of hip fracture was identical to that between the ‘post-osteoporotic vertebral compression fracture following initial treatment decision’ and the ‘patient sustains an additional vertebral fracture and remains alive’ states.

To ‘death through fracture or non-fracture related causes’

The methodology for calculating the risk of mortality from fracture and non-fracture causes is identical to that from the ‘post-osteoporotic vertebral compression fracture following initial treatment decision’ state.

To ‘patient sustains an additional vertebral fracture and a hip fracture and remains alive’

It was assumed that the risk of vertebral fracture for patients was independent of whether or not the patient was simulated to have sustained a hip fracture. Therefore, the methodology for calculating the risk of hip fracture was identical to that between the ‘post-osteoporotic vertebral compression fracture following initial treatment decision’ and the ‘patient sustains an additional vertebral fracture and remains alive’ states.

To ‘death through fracture or non-fracture related causes’

The methodology for calculating the risk of mortality from fracture and non-fracture causes is identical to that from the ‘post-osteoporotic vertebral compression fracture following initial treatment decision’ state.

To ‘death through fracture or non-fracture related causes’

The methodology for calculating the risk of mortality from fracture and non-fracture causes is identical to that from the ‘post-osteoporotic vertebral compression fracture following initial treatment decision’ state.