Percutaneous vertebroplasty and percutaneous balloon kyphoplasty for treating osteoporotic vertebral compression fractures

The outcomes of the Buchbinder and INVEST studies included pain measured on either a numeric rating scale or a visual analogue scale (VAS). The Buchbinder study reported no statistically significant differences in the change from baseline between vertebroplasty and operative placebo with injected local anaesthesia at 1 week, or at 1, 3 or 6 months for the primary outcome of overall pain, with a mean difference adjusted for stratification variables and baselines values of 0.7 (95% confidence interval [CI] -0.4 to 1.8) at 1 week, -0.5 (95% CI -1.7 to 0.8) at 1 month, -0.6 (95% CI -1.8 to 0.7) at 3 months and -0.1 (95% CI -1.4 to 1.2) at 6 months (negative numbers indicate less severe pain). The INVEST study reported no statistically significant differences in the change in pain from baseline between groups for overall pain at 3 days, 1 week and 1 month, with an adjusted mean difference of 0.4 (95% CI -0.5 to 1.5, p=0.37) at 3 days, 0.1 (95% CI -0.8 to 1.1, p=0.77) at 1 week, and -0.7 (95%CI -1.7 to 0.3, p=0.19) at 1 month. The INVEST study showed a clinically meaningful improvement in pain (that is, a decrease of 30% or more) with vertebroplasty at 1 month, but this effect was not statistically significantly different from operative placebo with local anaesthesia (64% compared with 48%; p=0.06). In addition, the Assessment Group's meta-analysis of the individual patient data from both studies found no statistically significant improvement in change in pain from baseline between groups at 1 month, with an adjusted mean difference of -0.6 (95% CI -1.4 to 0.2). However, the number of patients taking opioids for pain decreased over time in both groups in both studies. In the Assessment Group's meta-analysis, after adjusting for baseline opioid use, patients randomised to vertebroplasty were statistically significantly more likely to be taking opioids at 1 month than patients randomised to operative placebo with local anaesthesia (relative risk [RR] 1.25, 95% CI 1.14 to 1.36, p<0.001). Therefore, the Assessment Group stated that the trend towards a higher proportion of patients in the vertebroplasty group achieving an improvement of 30% or more in pain scores at 1 month may have been influenced by the fact that the vertebroplasty group was more likely to be using opioids than the operative placebo with local anaesthesia group.

The Buchbinder study also reported pain as an outcome in terms of QUALEFFO (Quality of Life Questionnaire of the European Foundation for Osteoporosis) pain scores and found no statistically significant differences between groups. Data were also collected on perceived pain, classified as 'better' or 'worse'. There were no statistically significant differences in the proportion of patients in each category at any time point. The INVEST study reported on the frequency with which patients experienced pain, and the impact of pain on their daily lives. For vertebroplasty and operative placebo with local anaesthesia, both pain frequency and pain 'bothersomeness' decreased between baseline and 1 month, with point estimates favouring vertebroplasty. However, the difference between vertebroplasty and operative placebo with local anaesthesia was not statistically significant.

The Buchbinder study presented health-related quality-of-life results based on AQoL (Assessment of Quality of Life), EQ-5D and QUALEFFO measures. The INVEST study presented health-related quality-of-life results based on EQ-5D and SF-36. AQoL scores were not different for vertebroplasty and operative placebo with local anaesthesia. EQ-5D scores, available in the Buchbinder study for 79% of patients in the vertebroplasty group and 73% in the operative placebo with local anaesthesia group, were not statistically significant different between groups for short- or medium-term outcomes. The Assessment Group's meta-analysis of individual patient data at 1 month also indicated that the result (with positive numbers indicating better quality of life) was not statistically significant (adjusted mean difference 0.03, 95% CI -0.02 to 0.08). The Assessment Group highlighted that, because 0.08 is the minimum clinically important difference for back pain on the EQ-5D scale, the confidence interval for the pooled data only just included the possibility of a clinically important difference favouring vertebroplasty. Based on QUALEFFO scores in the Buchbinder study, the only statistically significant result was at 1 week, with an adjusted mean difference of -4.0 (95% CI -7.8 to -0.2), but the Assessment Group stated that, because no minimum clinically important difference had been proposed for the QUALEFFO, the clinical significance of this result is not clear. The INVEST study found no statistically significant differences between treatment groups at any point using SF-36 scores.

Both the INVEST and Buchbinder studies assessed back-specific functional status using the modified 23-point version of the Roland-Morris Disability Questionnaire (RDQ). However, neither study showed any statistically significant differences for outcomes in the short-term (3 days to 2 weeks) or in the medium-term (1 month to 6 months). In addition, the Assessment Group's meta-analysis of individual patient data from both studies indicated no statistically significant difference between treatment groups at 1 month in terms of mean RDQ scores, with an adjusted mean difference of -0.8 (95% CI -0.9 to 2.4). The INVEST study included a post hoc analysis to identify the proportion of patients who achieved a clinically meaningful (although not defined) improvement in physical disability related to back pain at 1 month. There was no difference between the proportion of patients in each group who achieved a clinically meaningful improvement (40% of the vertebroplasty group and 41% of the operative placebo with local anaesthesia group, p=0.99). The Assessment Group's meta-analysis also found no statistically significant difference in RDQ scores in the proportion of patients improving by at least 3 units or by at least 30%. The INVEST study reported mean Study of Osteoporotic Fractures–Activities of Daily Living (SOF-ADL) scores at baseline and 1 month, with no statistically significant difference between treatment groups in change from baseline with an adjusted mean difference of 0.4 (95% CI -0.8 to 1.6, p=0.51).

All studies comparing vertebroplasty with optimal pain management reported pain measured on a numeric rating scale or VAS. The Farrokhi, Rousing and VERTOS II studies showed statistically significant improvements between groups in short- and medium-term changes from baseline in pain after vertebroplasty. However, the Assessment Group highlighted that the favourable result reported by Rousing at 1 month may have been unreliable because these data were collected almost a year after the event. The VERTOS II and Farrokhi studies also found statistically significant improvements in the change from baseline between groups in longer-term outcomes. However, the Assessment Group noted that in the VERTOS II study, when defining a minimum clinically important improvement as 2 or more points, the 95% confidence interval included the possibility that the results were not clinically meaningful. In the study by Blasco, statistical significance in change in pain from baseline was reported at 2 months, when the result favoured vertebroplasty.

The proportion of people taking opioids was not statistically significantly different between treatment groups in the Blasco study. However, the Assessment Group found the results from this study difficult to interpret, partly because a higher proportion of patients in the vertebroplasty group needed opioids at baseline. In the VERTOS study, patients in the vertebroplasty group used less analgesia and patients in the control group used more analgesia, resulting in statistically significant differences that favoured vertebroplasty. In VERTOS II, analgesic use fell in the vertebroplasty group compared with the control group at 1 day (p<0.001), 1 week (p<0.001), and 1 month (p=0.033), but not thereafter; the Assessment Group highlighted that the actual figures were not presented.

Both the Farrokhi and Blasco studies reported changes in height and deformity of the vertebral body, but the results cannot be compared because it is not clear whether the studies measured height by the same methods. The Blasco study showed no statistically significant or clinically important differences between treatment groups in the change in vertebral body height from baseline at 12 months. In contrast, the Farrokhi study showed that vertebroplasty statistically significantly improved mean vertebral body height throughout the first year but not thereafter, and statistically significantly improved and sustained angular deformity throughout the 36-month follow up period. The VERTOS II study reported data relating to the progression of treated fractures during follow up. At the last follow‑up (median 12.0 months, range 1 to 24 months), statistically significant moderate or severe height loss was seen in 12% of patients in the vertebroplasty group, compared with 41% of patients in the optimal pain management group (p<0.001).

The Rousing study assessed health-related quality of life using the Dallas Pain Questionnaire, which evaluates the impact of chronic pain on a patient's life. Only the score for work and leisure at 3 months reached statistical significance, favouring vertebroplasty compared with optimal pain management, but the score was unadjusted for the difference in scores at baseline. Comparing changes from baseline in each group, rather than directly comparing pain scores at 2 weeks, favoured conservative management.

VERTOS II and the Rousing study reported health-related quality of life using EQ-5D. The Rousing study provided EQ-5D utility values for 58% of patients in the vertebroplasty group and 71% in the optimal pain management group. The results, with negative differences indicating a worse outcome with vertebroplasty, indicated a mean group difference of -0.085 (95% CI -0.15 to -0.02) at 3 months and -0.169 (95% CI -0.23 to -0.11) at 12 months, compared with baseline. VERTOS II collected EQ-5D data throughout the study but reported only baseline values.

Blasco, VERTOS and VERTOS II reported health-related quality of life using QUALEFFO. Results from the Blasco trial indicated that, in the short and medium term, there was a non‑statistically significant improvement in scores with vertebroplasty compared with optimal pain management at all time points. The VERTOS study found that vertebroplasty was associated with better short-term total QUALEFFO scores than optimal pain management. In VERTOS II, after adjusting for baseline differences, there was a statistically significant difference in QUALEFFO scores at 1 year that favoured vertebroplasty (p<0.0001); however, actual scores were not reported. The Rousing trial reported health-related quality of life using SF-36 and showed no statistically significant differences between treatment groups at any point.

The VERTOS and VERTOS II studies used the RDQ (24-point version) to assess back-specific functional status. The VERTOS study reported that the between-group change from baseline to 2 weeks favoured vertebroplasty over optimal pain management, but the Assessment Group could not calculate the statistical significance because the study reported no measure of variability. The VERTOS II study reported a statistically significant difference that favoured vertebroplasty at 1 year compared with optimal pain management (p<0.0001); however, the study did not provide the RDQ scores or indicate what difference would reflect a clinically important difference.

The Farrokhi study reported disability using a modified Oswestry Disability Index (ODI) and reported that vertebroplasty was associated with a statistically significant improvement in change from baseline at all times from 1 week to 36 months compared with optimal pain management. The Assessment Group noted that, because 4 points on the ODI is the minimum clinically important difference for back pain, these differences were clinically meaningful throughout. The Farrokhi study also noted that all 40 patients in the vertebroplasty group could walk 1 day after vertebroplasty, but only 1 of the 42 patients (2%) in the optimal pain management group could walk 1 day after optimal pain management, indicating a relative risk of 28.3 (95% CI 5.9 to 136.5, p<0.0001).

The Rousing study reported functional outcomes using the Barthel Index, which provided data for a subset of the study population. At 12 months, the absolute score was statistically significantly better in the vertebroplasty group than in the optimal pain management group, but the difference between groups was no longer statistically significant when adjusted for differences at baseline. The Assessment Group stated that the result may indicate a ceiling effect, whereby there is little scope for vertebroplasty to improve functional outcome more than optimal pain management does. The Rousing study also reported 3 tests of physical function for a subset of the population: tandem, timed up and go, and repeated chair tests. No statistically significant differences between groups were noted at 3 or 12 months but, because the trial provided no baseline values, the change from baseline is not known.

The Liu study was the only study to compare vertebroplasty with kyphoplasty, and assessed pain, vertebral body height and angular deformity. It did not assess health-related quality of life. It assessed pain measured on a VAS and reported no statistically significant differences between vertebroplasty and kyphoplasty in the short or medium term, with a mean difference of -0.2 (95% CI -0.43 to 0.03) at 3 days and 0.1 (95% CI -0.28 to 0.48) at 6 months compared with baseline, with negative differences favouring vertebroplasty. However, the Assessment Group highlighted that the trial was likely to have been underpowered. For changes in vertebral body height and angular deformity, the trial reported that kyphoplasty led to statistically significantly greater improvements in both postoperative vertebral body height and angular deformity than did vertebroplasty, but the Assessment Group was not clear at what time points the study measured these outcomes.

The FREE study was the only study to compare kyphoplasty with a non‑operative treatment. For assessing pain as an outcome, the FREE study used SF-36. The results indicated that patients randomised to kyphoplasty improved more than those randomised to optimal pain management, the difference over a period of 12 months being 9.2 points (95% CI 3.9 to 14.6, p=0.0008). The FREE study also reported pain measured on a numeric rating scale and reported statistically significant long-term differences between groups; however, the Assessment Group noted that these differences were unlikely to reflect a clinically meaningful difference.

The FREE study reported the use of analgesics. Kyphoplasty statistically significantly reduced the need for opioid medication at 1 month and 6 months, but not at 12 months or 24 months. The Assessment Group highlighted that the FREE study did not report changes in vertebral body height, even though maintenance of vertebral body height was a secondary outcome. The study protocol stated that vertebral body height was measured only in patients having kyphoplasty, making comparison with optimal pain management impossible. The study did show a statistically significant improvement from baseline with kyphoplasty in the kyphotic angle of the fracture at 24 months, but the Assessment Group noted that the clinical significance of this result is not clear.

The FREE study used EQ-5D and SF-36 to assess health-related quality of life. Using EQ-5D, statistically significant differences in outcomes favouring kyphoplasty over optimal pain management were reported at 1, 12, and 24 months. However, the Assessment Group highlighted that, at a minimum clinically important difference for back pain of 0.08, the confidence intervals at 3, 6, 12 and 24 months included the possibility of effects that are not considered clinically important. Using the SF-36 physical component summary score, the study reported a statistically significant mean difference of 5.2 (95% CI 2.9 to 7.4, p<0.0001) between groups at 1 month, favouring kyphoplasty. Although the results remained statistically significant at 3 months and 6 months, the confidence intervals included the possibility of achieving a result that may not be considered clinically important and, after 6 months, there was no statistically significant difference between treatment groups. The FREE study also reported psychological wellbeing, which was assessed by the SF-36 mental component summary score, and identified no statistically significant differences between treatment groups, although the confidence intervals included the possibility of potential clinically important treatment effects favouring kyphoplasty compared with optimal pain management at time points up to 12 months.

The FREE study assessed back-specific functional status using the original 24-point version of the RDQ. It showed that kyphoplasty was associated with statistically significantly better outcomes compared with optimal pain management at 1 and 12 months, but not at 24 months. Moreover, at 12 months, the confidence intervals included the possibility of not achieving a clinically important outcome. The FREE study also reported that kyphoplasty was associated with a statistically significant reduction in the probability of needing walking aids at 1 month, but not at 12 months. However, the Assessment Group noted the data were not robust because of missing data. The FREE study also recorded the number of patients who reported 1 or more days of bed rest because of back pain in the previous 14 days. At 1 month, patients in the kyphoplasty group reported on average 2.9 fewer days of restricted activity than patients in the optimal pain management group (95% CI 1.3 to 4.6, p<0.001), but at 12 months the difference was no longer statistically significant (1.6 days, 95% CI -0.1 to 3.3, p=0.07). The actual numbers of days of restricted activity in each group were not reported.

The Assessment Group stated that the trials were not powered to determine differences in overall mortality, and noted that none of the studies showed any statistically significant differences in overall mortality between treatment groups. The Assessment Group also combined 12 months of mortality data from the Blasco, Rousing and VERTOS II studies comparing vertebroplasty with optimal pain management. The point estimate of the relative risk favoured vertebroplasty, but was not statistically significant (0.68; 95% CI 0.30 to 1.57, p=0.37). Medtronic provided a large observational study (n=858,979) based on US Medicare registry data with follow‑up to 4 years, which showed a mortality benefit with vertebroplasty or kyphoplasty compared with optimal pain management in patients with vertebral compression fractures, with a hazard ratio of 0.76 (95% CI 0.75 to 0.77) for vertebroplasty and 0.56 (95% CI 0.55 to 0.57) for kyphoplasty, adjusting for age, sex, race, Charlson comorbidity index and other coexisting diseases. In addition, kyphoplasty was associated with a greater mortality benefit compared with vertebroplasty, with an adjusted hazard ratio of 0.77 (95% CI 0.75 to 0.78; Edidin 2011). The Assessment Group stated that academic-in-confidence data provided by Medtronic on mortality at 5 years from US Medicare registry data, as well as data from a smaller observational study (n=3,607) based on a German health insurance fund, further supported a benefit in mortality associated with the technologies. The Assessment Group stated that, apart from the possibility of uncontrolled confounding, these studies raise the possibility that improvement in biomechanical factors after treatment improves survival.

The Assessment Group presented adverse events reported in the trials supplemented with observational studies and case reports.

All but the Liu and INVEST studies reported cement leaks confirmed by imaging, and all had used polymethylmethacrylate cement, presumed by the Assessment Group to be of low viscosity. The Blasco study found that cement leaks did not cause patients immediate complications. However, leaks into the inferior intervertebral disc increased the risk of incident vertebral fracture (odds ratio [OR] 7.2, 95% CI 1.7 to 69.3). The Farrokhi study reported 13 asymptomatic leaks and 1 symptomatic leak into the epidural space treated with urgent bilateral laminectomy. The Rousing study stated that none of the cement leaks caused neurological symptoms. In the VERTOS II study, most leaks were discal or into segmental veins, and cement pulmonary emboli were visible on computed tomography scan in 26% (95% CI 16% to 39%) of patients, although the patients did not have symptoms. In the FREE study, most leaks went into the vertebral end-plates or they were intervertebral disc leaks, with 1 leak into the vertebral foramina, no leaks into the spinal canal, and no cement emboli.

The Buchbinder, FREE and VERTOS II studies reported postoperative infections potentially related to treatment. In the Buchbinder study, investigators administered the intravenous antibiotic cephalothin prophylactically after cement injection. Osteomyelitis developed in a patient who did not receive an antibiotic because of allergies. In the FREE study, a patient developed spondylitis in the vertebral body 376 days after surgery. Sepsis or septic shock was reported in 1 patient in the kyphoplasty group and in 3 patients in the optimal pain management group. The Assessment Group also noted that 3 patients who underwent kyphoplasty subsequently had pulmonary emboli of venous origin, and the earliest of these occurred 46 days postoperatively. The Farrokhi study reported that no infections occurred, and the Rousing study reported that there were no adverse reactions other than cement leaks. In the remaining 4 studies (Blasco, INVEST, Liu, VERTOS), no postoperative infections were mentioned.

The risk of fracturing a vertebra adjacent to the treated vertebra was reported in 4 studies (Buchbinder, Farrokhi, FREE, Rousing), and none identified a statistically significant difference between treatment groups in the proportion of patients who experienced at least 1 clinically important fracture. However, the Blasco study noted that more (71%) of the radiographic fractures in the vertebroplasty group were clinically important compared with fewer (9%) in the optimal pain management group (OR 25.7, 95% CI 3.0 to 216.8, p=0.029); the investigators did not report the number of patients who had incident vertebral fractures.

The Assessment Group highlighted the potentially serious complications that can result from managing compression fractures conservatively. Bed rest can result in muscle wasting, deconditioning, deep vein thrombosis and pulmonary emboli. Opioid analgesics can cause undesirable adverse reactions including cognitive impairment, constipation and nausea, and non‑steroidal anti‑inflammatory drugs are associated with gastrointestinal and renal problems.

The Assessment Group stated that no trial data were identified for patients with or without fracture-related vertebral deformity or for inpatients at the time of randomisation. However, some data were available for subgroups based on the severity of pain at randomisation and for the time from fracture to intervention. No data for subgroups were available for kyphoplasty.

A meta-analysis by Staples et al. (2011) of individual patient data from the Buchbinder and INVEST studies grouped by baseline pain severity showed no statistically significant differences in RDQ scores, EQ-5D scores, or pain scores between patients with severe pain (score of 8 or more on a 0 to 10 scale) or mild-to-moderate pain (score of less than 8) at baseline. In both groups (vertebroplasty and operative placebo with local anaesthesia), patients with greater degrees of pain at baseline experienced a greater reduction in pain. The Assessment Group stated that this could reflect a greater potential for improvement. The Assessment Group also stated there were no data to suggest that outcomes would differ between patients who were or were not inpatients before treatment.

The INVEST study reported results by duration of pain at baseline in post hoc analyses and found no statistically significant difference between vertebroplasty and operative placebo with local anaesthesia on pain at 1 month, but was underpowered for this analysis. Data from the Staples study combining individual patient data from the INVEST and Buchbinder studies assessed the effectiveness of vertebroplasty in patients with fracture pain of recent onset (6 weeks or less) compared with pain of longer duration. Because the INVEST study allowed crossover after 1 month, the outcomes were compared up to that time point, finding no statistically significant differences in RDQ scores, EQ-5D scores, or pain scores.

Medtronic submitted a Markov tunnel model adapted from the Strom model (see section 4.2.1) to determine the cost effectiveness of kyphoplasty, vertebroplasty and optimal pain management in patients hospitalised with vertebral compression fractures. The model has a lifetime time horizon, 6-month cycles, and an NHS perspective. Costs and utilities are discounted at 3.5%. In the base case, Medtronic assumed that patients are 70 years old and have a T score of -3.0 SD, similar to patients in the FREE and VERTOS II trials. The model assumes people are either treated with kyphoplasty or an alternative, and remain in their initial treatment health state (progressing through the sub-states) until they die or experience another vertebral fracture that is treated using optimal pain management only. The manufacturer calculated the transition probabilities for further vertebral fractures from equations taking into account a patient's T score, age, number of previous fractures and, because the data were not available, the imputed ratio between the incidence of hip and vertebral fractures at each age, assuming that Swedish values (from Strom) apply to the UK. The transition probabilities to death use data from the Human Mortality Database for patients in the UK and the relative risks of mortality for people with a prior vertebral fracture (from Strom).

Medtronic took utility values for kyphoplasty and optimal pain management directly from the FREE trial, and for vertebroplasty indirectly from the VERTOS II trial, estimating values by adding the difference between vertebroplasty and optimal pain management to the scores for optimal pain management in the FREE trial. Because VERTOS II presented data on QALYs at baseline, 1 month and 12 months, Medtronic inferred the average utility across the 1-year period. Medtronic assumed that, unless a patient has a refracture, their utility will improve during the first 2 years and, thereafter, the utility in patients treated with optimal pain management will decline at the rate of the general population. For patients treated with kyphoplasty or vertebroplasty, the utility gain compared with optimal pain management declines linearly during the first year. Consequently after 3 years, unless patients have a refracture, they will have the same health utility, which declines at the same rate as the general population. The model assumes that both kyphoplasty and vertebroplasty improve survival compared with optimal pain management. The hazard ratios for death for kyphoplasty and for vertebroplasty are based on the US Medicare registry data. Medtronic included recurrent vertebral fracture but no other adverse events in the model, citing a lack of data as the reason, although acknowledging potentially substantial consequences of adverse events.

The list price of a kyphoplasty kit is £2,600.50, and the submission also noted an average selling price of £1,900. Medtronic assumed an acquisition cost of vertebroplasty that was commercial in confidence. Medtronic updated the costs of the preparatory, operating and postoperative phases from those in the Strom study. Medtronic obtained data on the length of stay in hospital after treatment from Hospital Episode Statistics 2010 to 2011 data, and the cost per day in hospital from NHS Reference costs 2009 to 2011. The deterministic analyses gave an ICER of £2,167 per QALY gained for kyphoplasty compared with optimal pain management and £2,053 per QALY gained for vertebroplasty compared with optimal pain management. The deterministic analysis of kyphoplasty compared with vertebroplasty gave an ICER of £2,510 per QALY gained, while probabilistic analyses gave ICERs of £2,118 (kyphoplasty compared with optimal pain management), £2,100 (vertebroplasty compared with optimal pain management) and £2,174 (kyphoplasty compared with vertebroplasty) per QALY gained.

Medtronic conducted sensitivity analyses to study the impact of changing 1 variable at a time: the time horizon; the discount rate for costs and QALYs; the health utility benefit from the FREE trial; the time at which the utility gain for kyphoplasty and vertebroplasty compared with optimal pain management is offset linearly; mortality rates after a fracture; the price of vertebroplasty compared with kyphoplasty; the unit costs per day in hospital; the assumed T‑score of the cohort; the age of the cohort; whether patients are treated with a bisphosphonate; and the proportion of patients who are male. The ICERs for kyphoplasty compared with optimal pain management and kyphoplasty compared with vertebroplasty remained below £15,000 per QALY gained in all instances. The assumption that vertebroplasty and kyphoplasty cause patients to live longer (greater for kyphoplasty than for vertebroplasty) was a key driver of the cost-effectiveness results and, when the manufacturer assumed no mortality benefit with either vertebroplasty or kyphoplasty, then the ICER for kyphoplasty compared with vertebroplasty was £27,340 per QALY gained. Medtronic noted that the ICERs for both vertebroplasty and kyphoplasty compared with optimal pain management remained low. The sensitivity analysis for changing the length of hospital stay after kyphoplasty also increased the ICER for kyphoplasty compared with vertebroplasty to over £20,000 per QALY gained.

Johnson and Johnson's model aimed to determine the cost effectiveness of vertebroplasty, kyphoplasty, optimal pain management and of operative placebo with local anaesthesia using a scenario analysis. The manufacturer developed a treatment-state model with a 1-year time horizon, an NHS perspective, and costs and benefits discounted at 3.5%. To estimate effectiveness measured by pain experienced by patients, Johnson and Johnson performed a network meta-analysis using VAS scores and EQ-5D data from trials. The manufacturer also performed an analysis based on a 'target population', that is, patients with fractures that occur within 3 months who are expected to benefit most from vertebroplasty or kyphoplasty.

The model has patients receiving different treatments and assigns values for pain using a VAS score before treatment and then again at 2 weeks depending on the intervention received. The model updates the treatment-dependent VAS score at 1 month, 6 months and 12 months. Johnson and Johnson used a regression analysis to describe the relationship between VAS and EQ-5D, based on data for both outcomes derived from a network meta-analysis. This relationship then allowed the manufacturer to model changes in quality of life from VAS scores that had been reported in trials at multiple time points. The Assessment Group stated that the manufacturer did not attempt in its network meta-analysis to extrapolate or interpolate data from trials that did not report VAS scores at the designated time intervals, and this could be a source of uncertainty when modelling pain scores over time. The Blasco study was published after completion of the manufacturer's systematic review. If the manufacturer had included this trial, which had similar VAS scores for vertebroplasty and optimal pain management (with both values being relatively high), the VAS scores in the model for all treatments would have increased and the relative difference between optimal pain management and both vertebroplasty and kyphoplasty would have diminished. Johnson and Johnson did model procedure-related adverse events.

Johnson and Johnson used a bottom-up costing approach based on published data from the Strom study for vertebroplasty and kyphoplasty. The manufacturer did not model costs for optimal pain management because it assumed that all patients, including patients receiving vertebroplasty or kyphoplasty, would receive optimal pain management. The manufacturer determined costs of the preparatory, operating and postoperative phases from the Strom study adjusted for inflation to 2009 to 2010 prices. The costs for vertebroplasty varied according to the number of vertebrae needing treatment, being £1,358 for 1 vertebra, £1,784 for 2, and £1,848 for 3. Based on Dr Foster data, the estimated frequency of treating 1, 2 or 3 fractures resulted in an average weighted cost of £1,472. The Assessment Group noted that, in its submission, Johnson and Johnson stated that 11 cm³ of cement was needed to treat 2 fractures but the manufacturer's calculations assume 7 cm³ of cement. If 11 cm³ were used, the average weighted cost would increase to £1,546. The cost of the kyphoplasty kit reported in the Strom study was adjusted for inflation by the manufacturer to a 2009 to 2010 cost of £2,842.

Johnson and Johnson's base-case results indicated that vertebroplasty was both more effective and less costly than kyphoplasty and therefore dominated kyphoplasty. The analysis of vertebroplasty compared with optimal pain management gave an ICER of £4,392 per QALY gained and kyphoplasty compared with optimal pain management gave an ICER of £14,643 per QALY gained. The results based on patients with fractures that occur within 3 months who are expected to benefit most from vertebroplasty or kyphoplasty also indicated that vertebroplasty dominated kyphoplasty; the analysis of vertebroplasty compared with optimal pain management gave an ICER of £4,755 per QALY gained and kyphoplasty compared with optimal pain management gave an ICER of £15,006 per QALY gained. The scenario analysis including operative placebo with local (injected) anaesthesia resulted in vertebroplasty dominating operative placebo with local anaesthesia, and operative placebo with local anaesthesia dominating kyphoplasty. The comparison of optimal pain management with operative placebo with local anaesthesia gave an ICER of £4,853 per QALY gained.

Johnson and Johnson performed several other scenario analyses, pooling data from operative placebo with local anaesthesia with data from optimal pain management; extending the time horizon to beyond 1 year for both the base‑case and target population with recent fractures; using an alternative bottom-up costing methodology and payment-by-results tariff for both the base‑case and the target population; and using direct EQ-5D values for both the base‑case and target populations. Vertebroplasty dominated kyphoplasty in all scenarios. The ICER for vertebroplasty compared with optimal pain management ranged from £568 to £13,595 per QALY gained in the base case and from £2,550 to £16,497 per QALY gained in the target population.

Johnson and Johnson performed univariate sensitivity analyses comparing vertebroplasty with optimal pain management and vertebroplasty with kyphoplasty. The main drivers of cost effectiveness were the efficacy of the treatment (that is, the VAS score at various time points), and costs (driven by the length of stay, cost per bed day and surgical equipment costs) for both the base case and the population with recent fractures. Results from probabilistic sensitivity analyses (varying parameters simultaneously) were broadly similar to the deterministic results in the base-case analysis; vertebroplasty dominated kyphoplasty for both base-case and target-population analyses. In addition, in the base-case analysis, the probabilistic ICER for vertebroplasty compared with optimal pain management in the model estimated the ICER at £4,388 per QALY gained in the base-case analysis and £4,711 per QALY gained in the target population analysis. The model estimated the probabilistic ICER for kyphoplasty compared with optimal pain management at £14,718 per QALY gained in the base-case analysis and £15,010 per QALY gained in the target population analysis. In addition, the Assessment Group corrected an error in the manufacturer's mathematical model in which only 10% of patients receiving kyphoplasty consume operating-room resources; the Assessment Group assumed that this value was intended to be 100%.

The Assessment Group's model was designed to determine the cost effectiveness of vertebroplasty, kyphoplasty, optimal pain management and operative placebo with local anaesthesia. The Assessment Group presented 6 scenarios rather than a base case. The Assessment Group stated that, given the uncertainty around whether vertebroplasty or kyphoplasty prolonged life, it organised results into 3 categories based on whether:

• kyphoplasty prolongs life more than vertebroplasty, which prolongs life more than optimal pain management

• vertebroplasty and kyphoplasty prolong life more than optimal pain management and

• vertebroplasty and kyphoplasty do not prolong life more than optimal pain management.
  
  The Assessment Group also stated that the results differed based on whether it took EQ-5D directly from the trials (INVEST, FREE and Buchbinder) or mapped stable VAS scores (which the Assessment Group defined as VAS scores assumed to occur at 1 month after operation for vertebroplasty or kyphoplasty, and at 3 months after optimal pain management treatment) to EQ-5D. In addition, the Assessment Group produced exploratory analyses assuming the use of high-viscosity cement.

The model consisted of 5 health states:

• the starting state of post-osteoporotic vertebral compression fractures when patients receive kyphoplasty, vertebroplasty, operative placebo with local anaesthesia, or optimal pain management

• a state in which a patient may experience a subsequent vertebral fracture

• a state in which a patient may experience a subsequent hip fracture

• a state in which a patient may experience both a subsequent vertebral and a hip fracture and

• death.
  
  The model allowed a patient to experience only 1 further vertebral fracture and 1 hip fracture. The Assessment Group assumed that a time horizon of 50 years reflects patients' lifetimes and the model employed 36 monthly time cycles followed by 47 yearly time cycles. The Assessment Group's rationale for the different cycle length was that different procedures may lead to different utilities in the period after a procedure, a difference more easily incorporated using monthly time cycles. Both costs and benefits were discounted at 3.5% per year. The model did not include the potential disutility associated with anxiety about the prospect of future fractures, or the potential reduction in bone mineral density associated with prolonged bed rest.

The Assessment Group estimated transition probabilities between health states from the literature. Taking into consideration that a patient's bone density is likely to decrease over time, the Assessment Group incorporated a decrease of 0.255 SD per 5‑year age group, assuming that women and men with the same T‑score have the same risks of fracture. If a patient was assumed to be taking a bisphosphonate, the assumed effect on vertebral fractures was based on relative risks reported in the literature. This effect was assumed to last for 5 years, with a linear decline in effect over a 5‑year period, so that the relative risk was 1 after 10 years. The risk of hip fracture or vertebral fracture was assumed to be independent of whether the patient was simulated to have a subsequent vertebral fracture or hip fracture.

The Assessment Group estimated the mortality rate associated with hip fracture from Stevenson et al. (2009) and the mortality rate associated with vertebral fracture from a UK study (Jalava et al. 2003). The model assumed that patients are more likely to die in the year in which a subsequent fracture occurs than they are thereafter. The model also assumed that the mortality rate after hip fracture must be equal to or greater than the mortality rates associated with a vertebral fracture in the age- and sex-matched general population. The mortality rate from causes other than fracture was taken from life tables from the Office for National Statistics, and the Assessment Group assumed that all patients die before they reach 101 years. When the Assessment Group assumed that patients who undergo kyphoplasty, vertebroplasty or operative placebo with local anaesthesia live longer than those who receive optimal pain management, mortality benefits were incorporated in the model for a period of 5 years in the base case. It was assumed that mortality benefits would cease immediately after 5 years. The Assessment Group assumed that the relative risks associated with treatment applied to all‑cause mortality and to the mortality rate associated with vertebral fractures, but not to hip fractures.

The Assessment Group calculated the hazard ratios within the 3 scenarios used to explore the effects of mortality using US Medicare registry data provided, academic in confidence, by Medtronic. The Assessment Group did not have data about any potential effect of operative placebo with local anaesthesia relative to optimal pain management on mortality, but assumed that the effect is half that observed for vertebroplasty because the effect of operative placebo with local anaesthesia on pain (VAS) was half that observed for vertebroplasty.

The Assessment Group assumed that utility values for all health states are a function of: sex; age; which procedure a patient undertakes; the time since the procedure; the time after which the model assumes that the utility values of patients treated with optimal pain management equals those of patients treated with an active intervention; the value of disutility after vertebral fractures that occurred more than 1 year before an intervention; and the mapping of VAS scores onto the EQ-5D. In addition, in the health states in which a patient had an additional vertebral and/or a new hip fracture, the model included a decrease in the utility value reflecting the fracture and, in the following cycles, persistent pain. The Assessment Group's model assumed that adverse events did not increase costs or disutility. However, the Assessment Group conducted a sensitivity analysis assuming that adverse effects led to QALY losses of 0.02 for kyphoplasty and vertebroplasty.

Costs within each of the health states were taken largely from the Stevenson study and adjusted for inflation to 2010 to 2011 prices using the Hospital and Community Health Services inflation indices. The Assessment Group took the cost of the high-viscosity Confidence Spinal Cement System from the Johnson and Johnson submission, although it assumed that 7 cm³ of cement was needed to treat 2 vertebral fractures, rather than 11 cm³. This gave an average cost of £1,546 per operation. The average estimated value for low-viscosity cement was £697. A clinical specialist advised the Assessment Group that approximately 15% of procedures would use high-viscosity cement or other more expensive cement types. The Assessment Group assumed that these more complex cases would add slightly over £100 to the average cost of an operation, resulting in an assumed cost of £800 per vertebroplasty procedure using low-viscosity cement. In calculating the ICERs, the Assessment Group assumed that vertebroplasty uses low-viscosity cement.

The Assessment Group adjusted the list price of £2,600.50 per kit for kyphoplasty to acknowledge that a proportion of patients would need kyphoplasty on more than 1 vertebra, which would require an additional pack of Kyphon HV-R bone cement, priced at £62 per pack. This resulted in the average price per patient increasing to £2,639 for kyphoplasty. The Assessment Group assumed that the cost of operative placebo with local anaesthesia was equal to vertebroplasty, but varied this assumption in sensitivity analyses.

The Assessment Group took costs for all phases of vertebroplasty and kyphoplasty from Johnson and Johnson's submission, estimated to be £540 for the preparatory, £243 for the postoperative and £528 for the operating phases. The Assessment Group chose costs for length of hospital stay data from Medtronic's submission, which used hospital episode statistics data, and chose the value for cost per hospital day of £232 from the Johnson and Johnson submission, noting that this value is an underestimate. The clinical advisers to the Assessment Group stated that most procedures would be performed as day cases and that length of stay would be shorter than suggested by hospital episode statistics data.

The Assessment Group performed sensitivity analysis for each scenario, exploring the impact of changes to the following assumptions: assuming a bed day cost of £0; changing the assumed cost of equipment for operative placebo with local anaesthesia and the cost of the procedure; changing the time of convergence (the point at which the pain score in patients undergoing vertebroplasty equals the pain score in patients receiving optimal pain management); and including potential QALY losses associated with adverse events.

The Assessment Group summarised that, in scenarios in which the model assumes that patients who undergo kyphoplasty live longer than those who undergo vertebroplasty, results indicated that kyphoplasty provided the most QALYs and gave ICERs below £12,000 per QALY gained, irrespective of whether the utility gain expressed in EQ-5D had been estimated by mapping stable VAS, or measured directly in the trials and even if the cost of kyphoplasty was increased, assuming a separate kit was needed for each level. The ICER for vertebroplasty compared with optimal pain management, when utility gain was estimated directly from EQ-5D in the trials, remained below £7,000 per QALY gained, except in 1 instance when it was extendedly dominated, when treatment benefit was assumed to disappear between 12 months and 24 months and the EQ-5D data from the Buchbinder trial were used.

In scenarios in which the model assumed that patients who undergo vertebroplasty and kyphoplasty live longer (but by the same degree) than patients who receive optimal pain management, and when the model assumes that patients who receive operative placebo with local anaesthesia also live longer, but only to half the degree as vertebroplasty and kyphoplasty, results indicated that vertebroplasty dominated kyphoplasty because it effectively provided the same QALYs at a higher cost. The ICER for vertebroplasty compared with optimal pain management remained below £10,000 per QALY gained across all assumptions except for the combination of assumptions in which: operative placebo with local anaesthesia was assumed to have an identical mortality benefit to balloon kyphoplasty and vertebroplasty; operative placebo with local anaesthesia was assumed to have a lower cost than vertebroplasty; adverse events for vertebroplasty were included; and the EQ-5D data from the randomised controlled trials were used. In this instance, vertebroplasty was dominated by operative placebo with local anaesthesia. However, it was noted that, if operative placebo with local anaesthesia was not seen to be an appropriate comparator, the ICER of vertebroplasty compared with optimal pain management remained below £10,000 per QALY gained.

In the scenarios in which the model assumed that patients who undergo vertebroplasty and kyphoplasty do not live longer than patients who receive optimal pain management, the cost-effectiveness results depend on whether the utility gain is estimated by mapping, but vertebroplasty nonetheless typically provided the most QALYs, and the ICER remained below £16,000 per QALY gained. The exception to this was when the Assessment Group adopted assumptions unfavourable to vertebroplasty, such as hospitalisation stay costs set at £0, reduced cost of operative placebo with local anaesthesia, incorporating adverse events for vertebroplasty, and an earlier convergence over time of EQ-5D scores. When the model estimated utility gained directly from the Buchbinder and INVEST trials, vertebroplasty always dominated kyphoplasty. Vertebroplasty was dominated by operative placebo with local anaesthesia in some cases and had an ICER greater than £20,000 per QALY gained in other cases. If the Assessment Group did not consider operative placebo with local anaesthesia as an appropriate comparator, vertebroplasty compared with optimal pain management had an ICER greater than £20,000 per QALY gained in some cases.

The Assessment Group also conducted an exploratory analysis assuming the use of high-viscosity cement for all patients. It stated that, for the cost per QALY gained to be equal to £20,000 per QALY gained, an additional 0.037 QALYs would be needed, a value greater than the value of 0.02 discounted QALYs assumed in the sensitivity analyses. The Assessment Group stated that it was unlikely that the ICER for high-viscosity cement compared with low-viscosity cement would be lower than £20,000 per QALY gained. However, the Assessment Group stated that a patient might need another operation if there was a problem with low-viscosity cement. So the Assessment Group estimated that, if more than 25% of patients needed another procedure on the same vertebra, then a strategy of using high-viscosity cement in all patients for the first procedure would be more cost effective. The Assessment Group stated that it was unlikely that a strategy of using high-viscosity cement in all patients rather than a subset selected by the clinician would have an ICER of less than £20,000 per QALY gained.