The clinical and cost effectiveness of ablative therapies in the management of liver metastases: systematic review and economic evaluation

Cryoablation

Cryoablation has a longer history than some other local ablative therapies, but there is controversy regarding its use in the liver, including recent guidance suggesting that it be used in the context of clinical trials only. 28 Cryoablation destroys tissue by delivering tissue-lethal freeze–thaw cycles to the tissue via cryoprobes through which a cryogen (liquid nitrogen or, more frequently, argon gas) is circulated. 19,26,29 By insulating the probe shaft and delivery hoses, cooling is limited to the probe tip. Liquid nitrogen and argon gas are capable of producing temperatures of at least –100 °C and cellular death results from direct freezing of tissue between –20 and –40 °C. 26,29 Freezing potentially produces large ablation zones and allows clearer delineation of the margins around the metastases. The procedure has usually required general anaesthesia and laparotomy for probe placement;29 however, cryoprobes small enough to be used percutaneously have been developed in recent years. 20 The efficacy of liver cryoablation is unclear. 28

Complications potentially include infections and haemorrhage, and perioperative mortality from myocardial infarction, pulmonary embolus, respiratory failure and cryoshock syndrome (a systemic response consisting of marked thrombocytopenia leading to coagulopathies, acute respiratory distress syndrome-like syndrome and myoglobinuria). 28

Laser ablation

Laser ablation uses a laser source such as neodymium-doped yttrium aluminium garnet (Nd: YAG)19,29 to deliver high-energy light (wavelength between 800 and 1064 nm) into the tumour. The optical fibres are placed into the tumour through a percutaneously located needle and the emitted photons travel up to 12 mm through soft tissue producing heating and cell death. 20 The terms ‘direct’ or ‘interstitial’ are often reported to clarify that optical fibres are inserted directly into the tissue; however, it has been recommended that ‘laser ablation’ replaces terms such as ‘laser interstitial tumour therapy’, ‘laser coagulation therapy’ and ‘laser interstitial photocoagulation’. 19

To enlarge the area of necrosis multiple fibres can be inserted into the tumour at regularly spaced intervals. This allows up to a 6–7-cm area of necrosis when the fibres are simultaneously energised. Adaptations have also been made to the fibre tip to avoid localised overheating and charring which decreases photon penetration, such as water-cooled fibres. 20,29 Treatment times vary but may exceed 1 hour for a large (6–7 cm diameter) ablation. 29

Smaller tumours show greater success30 and 5-year survival rates of 26% and median survival rates of 22–41 months have been reported across varying tumours. 29,31–35 Complications include pain, segmental infarction, abscess, pleural infusion and tumour seeding. 29

It is thought that this technique has been largely replaced by others, although it may still be used outside the UK.

Microwave ablation

Microwave ablation refers to all electromagnetic methods of inducing tumour destruction by using devices with frequencies from 30 MHz to 30 GHz. 19 As with radiofrequency ablation, microwave ablation involves placement of an antenna directly into the target tumour, under imaging guidance. Thermal coagulation of tissues is caused when microwaves (between 915 MHz and 2.45 GHz) oscillate water molecules and impart tissue heating through inefficiencies in this process. 20,29 Dependent on frequency, and unlike radiofrequency ablation, the zone of active heating around the probe feedpoint is approximately 2 cm either side of the probe. Microwave ablation does not utilise retractable tines and the resulting ablation tends to be slightly more elliptical. 29 However, the treatment sessions are shorter than for radiofrequency ablation with a 2-cm ablation zone produced in 60 seconds with microwave therapy. Multiple overlapping ablations are used to treat larger volumes. 20

Microwave ablation can be undertaken percutaneously, with imaging guidance and confirmation of probe positioning, or through open or laparoscopic surgery. 36

Adverse events include abscess, bleeding, infection, pneumothorax, colonic perforation, fever, pain, tumour seeding, pleural effusion, and (rarely) bile duct injury. Theoretical complications might include deterioration in liver function, and adjacent organ damage to the kidney, lung or heart. 36

At present, most of the evidence for microwave ablation has been published in East Asia and concerns hepatocellular carcinoma rather than metastatic disease. 26 A recent study based on registry data from 18 international centres included 140 participants, 81% of whom were treated with microwave ablation alone. This study suggests short times to achieve tumour ablation and low morbidity and mortality rates. 37 Clinical experts have stated that microwave ablation using 2.45-GHz probes are being used in the UK and appear to produce more rounded ablation zones with fewer issues of ‘reflected power’ and thinner ablation zones than seen with devices at 915 MHz.

Interventional procedure guidance issued in 2011 by the National Institute for Health and Care Excellence (NICE) stated that ‘current evidence on microwave ablation for the treatment of liver metastases raises no major safety concerns. The evidence on efficacy is inadequate in quantity and quality. Therefore this procedure should only be used with special arrangements for clinical governance, consent and audit or research’. 36 The cost-effectiveness of microwave ablation was not discussed in the NICE guideline.

Radiofrequency ablation

The term radiofrequency ablation applies to coagulation induced from all electromagnetic energy sources with frequencies less than 30 MHz. 19 An electrode is inserted into the tumour, usually assisted using some form of image guidance. 19,20 High-frequency alternating current (460 kHz)20,29 is transmitted from the tip of the electrode into the immediate tissue, causing ionic agitation and subsequent frictional heating of localised tissue to 100 °C, resulting in tissue death in an approximately spheroid volume of tissue. 26,29,38 The zone of active heating is, however, within only a few millimetres of the probe tines and tissue destruction is considerably reliant on conductive heating with its vagaries in the in vivo setting. A 2-cm to 5-cm spherical thermal injury can be produced with each ablation, which takes in the region of 20 minutes. 29

The needle tip has to be kept as cool as possible to prevent the adjacent tissue becoming charred, which increases impedance and prevents conduction of current into the tumour beyond the zone of ablation. 20 There are many electrode modifications available, and the type used influences the extent and predictability of ablation. Multitined expandable electrodes have an array of multiple electrode tines that expand from a central needle,19 allowing distribution of current distant from the needle tip. 20 Internally cooled electrodes are perfused with saline or water to prevent overheating of the needle tip. Perfusion electrodes have small apertures at the active tip that allow saline to be infused or injected into the tissue before, during or after ablation. 19

Depending on the size, location and number of tumours, radiofrequency ablation can be undertaken percutaneously, laparoscopically, or through open surgery (laparotomy). 38 The percutaneous approach is the least invasive and minimises morbidity. It is used for a limited number of small tumours usually remote from hollow viscera. ‘Hydrodissection’ techniques can be utilised to displace adjacent thermally sensitive organs, for small recurrences after prior surgical resection and for patients who are not candidates for other approaches for anatomic or clinical reasons. The laparoscopic approach reduces access morbidity and can permit some mobilisation of the liver, and detection of additional hepatic or extrahepatic disease. The open laparotomy approach is the most invasive and can be performed alone or in conjunction with other procedures such as surgical resection. It allows fuller mobilisation of the liver and use of the Pringle manoeuvre, and is most often used for large tumours, larger number of tumour or tumours in difficult locations. 38

A systematic review of complications of intraoperative radiofrequency ablation of liver metastases39 reported overall mortality ranging from 0% to 1.8% in studies of people undergoing radiofrequency ablation without surgical resection, and morbidity rates ranging from 0% to 16%. The review reported rates of wound infection (0–1.8%), biliary complications (0–1.8%), pleural effusion (0–3.6%), liver failure (0–1.8%), vascular complications (0%) and skin burns (0–1.8%). Complications specific to radiofrequency ablation include hepatic abscesses, caused by infection of the necrotic tissue in the ablation site; biliary stenosis, which appear after the fibrous healing of biliary tract thermal damage; vascular thrombosis, caused by thermal endothelial damage; and skin burns, which occur when the dispersion surface is inadequate for the radiofrequency power.

Interventional procedure guidance issued in 2009 by NICE stated that ‘current evidence on the safety and efficacy of radiofrequency ablation for colorectal liver metastases is adequate to support the use of this procedure in people unfit or otherwise unsuitable for surgical resection, or in those who have previously had hepatic surgical resection, provided that normal arrangements are in place for clinical governance, consent and audit’. 40 The cost-effectiveness of radiofrequency ablation was not discussed in the NICE guideline.

Ultrasound ablation (focused ultrasound)

High-frequency sound waves (ultrasound), which use energy levels of 30–100 W, can heat tissue to over 90 °C and ablate liver metastases. Two methods of application of the ultrasound are currently used: extracorporeal (or transcutaneous), and direct, for percutaneous application with a needle-like applicator. 19,41 In the former method, the ultrasound ablation is truly non-invasive. Most reports of the use of ultrasound ablation to date relate to prostate, uterine fibroids and hepatocellular carcinoma, but it can be used in any solid tumour, including metastatic disease in the liver. Although early reports in hepatic solid tumours indicated some promise,42 clinical experts advise that it is not widely used in the UK.

Acetic acid ablation

An alternative chemical agent which can be used for chemical ablation is acetic acid19 (alternative descriptions include percutaneous acetic acid injection23). The procedure itself mirrors that of ablation with ethanol. Most reports to date relate to its use in hepatocellular carcinoma rather than liver metastases and clinical experts have indicated that it is not widely used in the UK.

Electrolytic ablation (electrolysis)24,26

Electrolytic ablation is performed by passing a direct current between two platinum electrodes. This creates a change in the pH of the surrounding tissues, with an acidic environment around the anode and an alkaline environment around the cathode. 24,26 Chemical reactions then also cause the creation of toxic products such as chlorine and hydrogen ions26 and these chemical changes result in tissue destruction in the tumour. There is, at the present time, a lack of high-quality published data on the efficacy of electrolytic ablation in liver metastases. 24 Clinical experts have stated that electrolytic ablation is not widely used in the UK.

Ethanol ablation

Percutaneous ethanol injection is the chemical ablative technique with the most extensive clinical experience. Most reports to date relate to its use in hepatocellular carcinoma rather than liver metastases. It is relatively simple to perform, quite inexpensive, and requires minimal equipment. Percutaneous ethanol injection is performed by the injection of absolute alcohol through a needle placed percutaneously directly into a tumour. The necrosis produced by ethanol injection results from cellular dehydration and tissue ischaemia from vascular thrombosis. However, ethanol ablation is not as effective as thermal ablation techniques for the treatment of metastases due to the limitations of adequate physical dispersal. For appropriately sized tumours, ethanol ablation improves survival and compares favourably with surgical resection for patients with hepatocellular carcinoma. 43,44

Chemoembolisation

Transarterial chemoembolisation provides a treatment option that is minimally invasive as a cytoreductive therapy. It is an alternative to systemic chemotherapy, surgical resection and non-surgical ablative techniques to treat resectable and non-resectable tumours. This technique aims to selectively target the liver metastases by delivering chemotherapy locally to the tumour via a hepatic arterial catheter combined with embolisation of the blood flow which acts both to prevent washout of the chemotherapeutic agent into the general circulation and to cause selective tumour ischaemia,45 the aim being to improve the pharmacodynamic availability of the delivered drug and reduce systemic toxicity. It is felt that liver metastases are particularly amenable to chemoembolisation given their relative arterialisation against background and the fact that the liver parenchyma receives two independent blood supplies, which serves to minimise normal parenchymal injury. Commonly used chemotherapeutic agents include doxorubicin, cisplatin and mitomycin (Mitomycin C Kyowa^®, Kyowa Hakko) which may be used either singly or in combination. The chemotherapeutic agent(s) may also be mixed with ethiodised oil (i.e. lipiodol), a viscous material that was originally used as a contrast agent but which also helps to localise the chemotherapy drugs inside the tumour cells. 45,46 A variety of different agents have been used to achieve embolisation and these agents have different properties. The properties of the particles that make up the embolic agent are an important consideration as these factors determine whether the embolisation will be permanent or transient, and what diameter of capillary will be occluded by the particles and the rate of drug elution from the embolic material. Infusion rate and particle size, shape (spherical or non-spherical) and concentration all affect the way that the particles penetrate the target tissue. 45,46 New controlled-release technologies in the form of drug-eluting beads provide a means to simultaneously deliver embolisation and chemotherapy. Furthermore, drug-eluting beads release the chemotherapeutic agent that they are loaded with in a more sustained manner for a prolonged period of time, thereby enhancing drug delivery to the tumour and reducing systemic toxicity. 45,46 In general, the systemic concentration of the chemotherapy drug being delivered via drug-eluting beads is low, which helps to limit possible systemic side effects.

Treatment may be repeated for the same tumour to enhance the adequacy of treatment and may be repeated in a staged approach in the contralateral lobe, for example, so as to reduce overall hepatotoxicity. A recent non-systematic review46 cites RCTs which showed increased overall survival for treating hepatocellular carcinoma. The evidence reported by this review indicates some benefits when treating hepatic metastases of breast, colorectal, and neuroendocrine primary tumours although none of the studies cited appears to be a RCT in which overall survival is reported. As drug-eluting bead transarterial chemoembolisation is a more recent technology, the evidence base is less mature; nevertheless, a recent systematic review47 identified eight studies. Three of these are case series focusing on liver metastases, which report response rates and procedure associated complications. No survival outcomes are reported.

Radioembolisation

Although radiotherapy is an important part of the treatment of most malignancies, its use in hepatic cancers has been limited due to the low tolerance of the organ to radiation, the risk of radiation hepatitis and damage to adjacent organs. 20 Recent technological developments (e.g. co-axial microcatheters and the development of embolic carrier particles of ∼30 μm) have, however, permitted targeted internal delivery of radiotherapy to metastases while minimising background irradiation, a form of ‘microbrachytherapy’. 48 As such, it relies on careful preparation of the liver to determine vascular anatomy and embolic exclusion of potential extrahepatic collaterals so as to ensure appropriate deployment of the microembolic to the target metastases. To ensure that no particles are able to travel to non-target sites, embolisation with metal coils20,21 of vessels that feed tissue beyond the area to be treated may also be required before treatment begins. The microspheres that act as the carrier particles may be made of glass, for example TheraSpheres^® (Nordion, Ottawa, ON, Canada) which are 20–30 μm in diameter,49 or resin, for example SIR-Spheres^® (Sirtex Medical Ltd, Sydney, NSW, Australia) which have diameters between 20 and 60 microns. 50 Both of the two commercially available microspheres contain the β-emitter yttrium-90 (⁹⁰Y), which has a maximum range of emissions in tissue of 11 mm (mean 2.5 mm) and a half-life of 64 hours. 49,50 Providing that the microspheres are correctly directed to the tumour, the short penetration distance means that radiation doses above 120 Gy can be delivered to the tumour with tolerable levels of radiation to the normal liver tissue. 48 Radioembolisation has been used in the treatment of primary and metastatic liver tumours. Unlike local ablation techniques, surgical resection and external beam radiation, the scope of the therapy is not limited by the number, size or location of the tumours within the liver. 20,48 However, there are some contraindications to treatment which include arterial-systemic shunting from liver to lungs and major arterial reflux from the hepatic vasculature to arteries supplying the gastroduodenal region. Radioembolisation is usually delivered with concomitant local or systemic chemotherapy. A recent meta-analysis51 identified two RCTs in which resin microspheres were used to treat colorectal metastases with the encouraging results of a survival benefit for the radioembolisation arm. 51 There is also evidence showing benefits in reducing liver metastases, allowing subsequent surgical resection. It is also thought to increase time to progression, have limited adverse effects, be tolerable and maintain quality of life. This study also included non-RCT data and studies focusing on hepatocellular carcinoma but a conclusion could not be drawn from this larger pool of evidence about whether glass or resin microspheres were more effective. Radioembolisation is relatively easy to deliver and can be undertaken as a day case or with a one-night stay, but requires significant angiographic work-up before administration.

Surgical resection of metastases

The aim of surgical resection of liver metastases is to remove all macroscopic disease with clear margins and leave sufficient functioning liver. 17 Five-year survival rates following the procedure range from 25% to 39%. 15,16

Selection criteria for liver resection usually include controlled or controllable primary tumour, no extrahepatic metastases detectable (or extrahepatic disease that can also be completely resected) and surgical resection technically feasible with tumour-free margins. Surgical resection would not be undertaken in people with such widespread hepatic involvement that residual liver function following surgical resection would be inadequate. People should be fit enough to tolerate general anaesthesia, have no major comorbidity and ideally have normal liver function. People with extrahepatic disease that should be considered for liver resection include resectable/ablatable pulmonary metastases; resectable/ablatable isolated extrahepatic sites, for example spleen, adrenal, or resectable local recurrence; and local direct extension of liver metastases to diaphragm/adrenal that can be resected. 52 In the case of liver metastases from colorectal cancer, approximately 20–30% of people may have disease that is potentially resectable. 17 A further 20–30% of people with unresectable liver limited disease might be brought to surgical resection with curative intent using induction systemic chemotherapy. 53 In some people curative surgical resection is aborted during surgery as metastases are unresectable.

Chemotherapy

Chemotherapy provides a clinically effective option for treating liver metastases in people who are not considered suitable for surgical resection. 6,54 As well as having been shown to improve survival itself, chemotherapy has resulted in the downsizing of tumours in cases where it was initially thought not possible to resect due to their location or inadequate hepatic functional reserve. 6,54,55 Where a compromised hepatic function reserve is a concern,56 preoperative chemotherapy may reduce overt tumour volume, so increasing the potential volume of future remnant liver. 57 In some cases the use of chemotherapy has allowed subsequent surgical resection or ablation, improving the chances of long-term survival and the possibility of cure. 58,59 Others have used a two-stage approach to surgical resection with an initial non-curative surgical resection or ablation followed by chemotherapy and further surgical resection to remove any remaining tumour. 59,60 Consideration of such strategies is recommended to be taken by a regional hepatobiliary unit.

Several options for chemotherapy have been recommended by NICE. 61,62 These include oxaliplatin-based regimens and irinotecan-based regimen for people with non-resectable liver metastases. 62 NICE have also recognised the benefits of the addition of therapies that target the epidermal growth factor receptor to conventional cytotoxic chemotherapies, which are thought to improve the chances of surgical resection of previously irresectable or suboptimally resectable tumours considerably, although they are of benefit only in patients who do not have activating mutations on the K-RAS gene. The addition of cetuximab (Erbitux,^® Merck Serono) to oxaliplatin-based regimens and irinotecan-based regimens (if unable to tolerate, or contraindications to, oxaliplatin) has also been recommended by NICE for first-line treatment of metastatic colorectal cancer. 61

Best supportive care

Best supportive care (BSC) is often used as the comparator arm in cancer trials; however, it is usually not well defined and may be left open to local interpretation. A Cochrane systematic review of supportive care for gastrointestinal cancer found that the descriptions of supportive care in the included studies were often vague and heterogeneous, making direct comparisons difficult. 63 BSC may include antibiotics to control infections, analgesics (including non-steroidal anti-inflammatory drugs and opioids), antiemetic drugs, transfusions for anaemia, corticosteroids, nutritional support, localised radiation therapy to alleviate symptoms such as pain, and psychosocial support. The European Organisation for Research and Treatment of Cancer (EORTC) Pain and Symptom Control Task Force agreed on the following definition: ‘supportive care for cancer patients is the multi professional attention to the individual’s overall physical, psychosocial, spiritual and cultural needs, and should be available at all stages of the illness, for patients of all ages, and regardless of the current intention of any anti-cancer treatment’. 63 Chemotherapy can be administered as a palliative treatment to people with unresectable metastatic disease. It can prolong both the time to tumour progression and survival. 6

Overall survival

Shibata and colleagues72 calculated overall survival (which was one of their two primary outcomes) from the time of treatment to the end of the follow-up period (the length of this was not reported). During follow-up nine of the microwave ablation group and 12 of the surgical resection group died (Table 10). There was no statistically significant difference in the estimated cumulative survival rates (p = 0.83). Mean overall survival in the microwave ablation group was 27 months versus 25 months in the surgical resection group. The estimated 1-, 2- and 3-year survival rates also indicated that therapeutic efficacy was similar between the two treatments (see Table 10).

Response

Response data were not reported by Shibata and colleagues. 72

Disease-free survival

Shibata and colleagues72 reported the outcome of disease-free survival but did not define this (Table 11). There was no statistically significant difference in the mean duration of disease-free survival between microwave ablation and surgical resection groups (11.3 months vs. 13.3 months respectively, p = 0.47).

Indications of surgical invasiveness

Surgical invasiveness was the second of the study’s primary outcomes. 72 Although there was no statistically significant difference between the two interventions for operation time or the length of hospitalisation there was statistically significantly less intraoperative blood loss in the microwave ablation group compared with the surgical resection group {mean 360 ml [standard deviation (SD) 230 ml] vs. 910 ml [SD 490 ml], p = 0.027}. Furthermore, no participants in the microwave ablation group required a blood transfusion, whereas six participants (38%) required a mean volume of 540 ml of transfused blood in the surgical resection group (Table 12).

Adverse events

Adverse events occurred in both trial arms with no statistically significant difference between them (two participants in the microwave ablation group vs. three participants in the surgical resection group, p = 0.87). In the microwave ablation group one participant had a hepatic abscess and one had a bile duct fistula. A bile duct fistula was also a complication for one participant in the surgical resection group, the other adverse events in the surgical resection group being an intestinal obstruction and a wound infection (Table 13).

Overall survival

Taguchi and colleagues81 calculated overall survival in all randomised participants when the majority of the participants in the treatment groups had died (number of deaths not reported). Although participants in the chemoembolisation arm (intra-arterial mitomycin C with degradable starch microspheres) had a longer median survival time than those receiving intra-arterial mitomycin C only, the difference between the groups was not statistically significant (9.7 months vs. 7.6 months, no p-value reported) (Table 22). Agarwala and colleagues82 reported that median overall survival for all treatment groups was 8.5 months with no data presented for the separate arms of this study. Consideration should be given to the uncertainties on key methodological attributes that were identified in the assessment of methodological quality when interpreting these results.

Response

The two RCTs reported on the response of the tumour to treatment, but used slightly different criteria for classifying response (see Appendix 7). Taguchi and colleagues81 reported that 42 of the 60 enrolled participants could be evaluated for tumour response and found that the chemoembolisation group had a higher response rate (complete response + partial response) than the group receiving intra-arterial mitomycin C only (54.5% vs. 20%, p < 0.05). Results for response were also provided separately according to primary tumour type (Table 23). In the RCT by Agarwala and colleagues,82 none of the participants achieved a complete response, and only 3 of the 17 evaluable participants achieved a partial response. The statistical comparisons carried out are not described in detail but the study authors state that there was no statistically significant difference in the response rates between the participants who received chemoembolisation and those who did not, nor was there any difference between the two chemotherapy dose tiers (p = 0.2769). When interpreting the results of these RCTs,81,82 consideration should be given to the uncertainties on key methodological attributes that were identified in the assessment of methodological quality.

Adverse events and toxicity

Pain, gastrointestinal disturbances and fever were experienced by a greater proportion of the participants who were treated by Taguchi and colleagues81 with chemoembolisation than by those treated with intra-arterial mitomycin C alone and the difference between the groups was statistically significant for all symptoms (Table 24). Five participants from each arm were withdrawn from the trial because of the adverse events experienced. The participants withdrawn from the group receiving intra-arterial mitomycin C alone were all withdrawn because of adverse results from clinical chemistry analysis. Two participants were also withdrawn for this reason from the chemoembolisation group, and the other three participants withdrawn from this group were withdrawn for subjective reasons (not further defined). In addition to the data reported here, the Taguchi and colleagues study81 reported on hepatocellular carcinoma. These data did not meet the inclusion criteria for the review. Taguchi and colleagues81 report two deaths that occurred during their study, but it is not clear whether these deaths were of participants with liver metastases or hepatocellular carcinoma.

Toxicity was the primary outcome of the study by Agarwala and colleagues,82 which aimed to establish the maximum tolerated dose of cisplatin with and without polyvinyl sponge. Dose-limiting toxicity did not occur at the 100 mg/m² cisplatin dose tier, but half of the participants who received cisplatin (either with or without polyvinyl sponge) at the 125 mg/m² dose tier experienced dose-limiting toxicity. A greater number of grade 4 toxicities were also experienced by the participants at the 125 mg/m² dose tier than participants at the 100 mg/m² dose tier. Across the whole study population the most common toxicity during treatment was a transient alteration in liver function tests (grade 1 or 2), with only one participant requiring dose reduction because of this. Haematological toxicity was less common than liver toxicity. Two participants developed anaemia (grade 1–2) and two developed thrombocytopenia (grade 3). It is assumed (although not clarified in the paper) that the two occurrences of thrombocytopenia are included in the earlier reporting of grade 3 toxicities. No statistical comparisons were reported between the groups but this may have been due to the small numbers of participants in each study arm of this Phase I/II RCT. The study authors concluded that their results indicated that the maximum tolerated dose of cisplatin for hepatic artery infusion (either with or without polyvinyl sponge) was 125 mg/m². Agarwala and colleagues82 noted that there were no complications related to the catheterisation procedure.

Overall survival

In both studies,83,84 survival was calculated from the commencement of the first treatment with chemoembolisation until death or last follow-up examination. In the study which included participants with various primary cancers,83 the median survival was reported to be 26.2 months (Table 27). In the study which included participants with metastatic breast cancer,84 the mean survival was reported to be 32.5 months (see Table 27). Survival rates at 1 2, 3 and 5 years were reported in one study only84 and were 88.8%, 55.9%, 36.6% and 13.7%, respectively.

Response

Vogl and colleagues84 reported on the response of the tumour to treatment as measured by the Response Evaluation in Solid Tumours (RECIST) criteria (see Appendix 7). The study reported the numbers of participants fulfilling the criteria for each category of response and results showed that 62 (38.5%) of participants had a complete response, 8 (5%) a partial response, 20 (12.4%) stable disease and 71 (44.1%) progressive disease. Of those with progressive disease, 64 (90%) underwent subsequent chemoembolisation treatment. Vogl and colleagues84 also reported the mean time to progression as being 8.2 (SD 12.29) months and overall tumour control as 13.1 (SD 15.9) months. The study also reports the proportion of participants with local recurrence of the lesion (see Appendix 4).

Adverse events: chemoembolisation followed by laser ablation

Complications and side effects of treatments were reported in both studies (Table 28). These are reported per participant, and show that adverse events varied, with many events being mild to moderate. Some more marked complications were reported, including in one study where six events required further intervention. 84 No deaths occurred in the 2011 study,84 but one death occurred within 30 days in the earlier Vogl and colleagues 200383 study.

The uncertain quality evidence from these studies should be taken into consideration when interpreting the results discussed above.

Overall survival

Estimated median overall survival was reported in all three included RCTs,85–87 with the RCT by Grey and colleagues85 also reporting estimated percentage survival at 1, 2, 3 and 5 years. In the two studies comparing radioembolisation with systemic chemotherapy, participants in the radioembolisation arm had a longer median survival than those receiving chemotherapy alone (Table 32). This difference was statistically significant in only one of these studies [hazard ratio (HR) 0.33; 95% CI 0.12 to 0.91; p = 0.025]. 87 In this study87 overall survival was calculated in all randomised participants when the majority of the participants in the treatment groups had died (with one participant in the radioembolisation group alive). This study was rated on a number of methodological attributes to have a low risk of bias, although the sample size was small and this should be considered when interpreting the results. In the Hendlisz and colleagues  study,86 which showed a non-statistically significant trend for longer median survival in the radioembolisation arm compared with the chemotherapy arm, the overall survival was calculated from time of randomisation until death from any cause. Seven participants remained alive at the time of the analysis; these participants had a median follow-up of 10 months, and this may in part account for the shorter duration of estimated survival seen in this study compared with the other two RCTs.

In the RCT85 comparing radioembolisation with hepatic artery chemotherapy, survival rates were calculated from the time of randomisation to death or last follow-up. In this study the majority of participants had died at the time of follow up (minimum of 3.5 years, at which time 65 of 70 participants had died). Median survival was reported as being 17 months in the radioembolisation group compared with 15.9 months in the hepatic artery chemotherapy group (HR 1.41, 95% CI 0.86 to 2.34, p = 0.18). Survival to 1, 2, 3 and 5 years was achieved by a higher percentage of those in the radioembolisation group than of those receiving hepatic artery chemotherapy alone (see Table 32). These differences were not statistically different between the two groups (p-value not reported). The trial authors state that the HRs suggest that those receiving hepatic artery chemotherapy alone had a 40% higher death rate than those receiving radioembolisation. In addition, the study authors stated that Cox regression suggests those treated with radioembolisation who survive more than 15 months experience a survival advantage compared with those treated with hepatic artery chemotherapy alone (p = 0.06). However, these results should be treated cautiously as it appears that this analysis was not defined a priori.

The results of these studies are not directly comparable, and each has some methodological shortcomings; however, results suggest that there may be a trend for a survival advantage following treatment with radioembolisation.

Response

The three included RCTs reported on the response of the tumour to treatment, but used slightly different criteria for classifying responses seen (for further details, see Appendix 7). In addition, Grey and colleagues85 measured response by two different means, by tumour area and by tumour volume, and reported data for both of these measures, and Van Hazel and colleagues87 reported response data from the ‘first integrated response’ and the ‘best confirmed response’. Grey and colleagues85 found that those in the radioembolisation group had a higher response rate (complete response and partial response) than the group receiving hepatic artery chemotherapy alone [44% vs. 18% (p = 0.01) when measured by tumour area, and 50% vs. 24% (p = 0.03) when measured by tumour volume]. Hendlisz and colleagues86 reported disease control rates as partial response and stable disease (no complete responses occurred). They found that the disease control rate in those in the radioembolisation group was higher than in those in the chemotherapy alone group [86% vs. 35%, p = 0.001 (95% CI for the difference between groups: 0.19 to 0.71)]. Both of these RCTs85,86 were rated as having a number of uncertainties on key methodological attributes which should be considered in the interpretation of these results. Van Hazel and colleagues87 did not report an overall estimate of tumour response rates in their study. 87

As can be seen in Table 33, the studies reported the numbers of participants fulfilling the criteria for each category of response. These were generally seen to be better in the radioembolisation arms compared with the comparator arms, although statistical significance testing was not undertaken in all cases. In the Grey and colleagues RCT85 when response was measured by tumour area, two participants in the radioembolisation arm achieved a complete response and 14 a partial response. In the hepatic artery chemotherapy arm there were no complete responses and six partial responses. In terms of progressive disease, rates were lower in the radioembolisation arm compared with the hepatic artery chemotherapy arm. Rates of participants with no change in their tumour status were, however, the same across both arms. Overall the groups were reported to be statistically significantly different on all measures of response (p = 0.01). Similar patterns of response were seen when measuring this outcome by tumour volume (see Table 33). The authors of this study also reported that one participant in each arm had disease reduced to such an extent that they were subsequently deemed surgically resectable and underwent surgical excision of metastases.

In the two studies comparing radioembolisation with systemic chemotherapy, similar patterns of response occurred, although no complete responses were observed in either study. In the RCT by Hendlisz and colleagues86 two participants achieved a partial response in the radioembolisation arm compared with no participants in the chemotherapy-alone arm, but this difference was not statistically significant (95% CI for the difference between groups –0.10 to 0.32, p = 0.22). The number of participants with stable disease was higher in the radioembolisation arm than in the chemotherapy arm, and the number of participants with progressive disease was lower in the radioembolisation arm than in the chemotherapy arm. 86 In the RCT by Van Hazel and colleagues87 there were eight participants whose best confirmed response was a partial response in the radioembolisation group compared with no participants in the chemotherapy-alone group. Best confirmed responses of stable disease and progressive disease were fewer in the participants treated with radioembolisation compared with those treated with chemotherapy (3 vs. 6 stable disease, and 0 vs. 4 progressive disease). Across all categories of response the groups were reported to be statistically significantly different (p = 0.001).

The results on response measures from the three studies are not directly comparable; however, they suggest that radioembolisation treatment leads to better tumour response than systemic chemotherapy or hepatic artery chemotherapy treatment alone.

Time to progression

All three included RCTs reported outcomes related to time to disease progression. Grey and colleagues85 and Hendlisz and colleagues86 reported the median time to progression in the liver. Hendlisz and colleagues86 additionally report the median time to progression (any), which is also reported as median time to progressive disease in the Van Hazel and colleagues’ study. 87 The study by Grey and colleagues85 measured progression by two different means, by tumour area and by tumour volume, and reported data for both of these measures, and Hendlisz and colleagues86 reported progression from ‘all progressions considered as events’ and ‘patients with treatment change censored at the time of change’. The results, seen in Table 34, are therefore not comparable, and for these studies results should also be interpreted in the context of some uncertainties around the potential for bias and/or small sample sizes.

Median time to progression in the liver was reported to be statistically significantly different between radioembolisation and hepatic artery chemotherapy in the Grey and colleagues85 trial when measured by either tumour area (15.9 months vs. 9.7 months, respectively, p = 0.001) or tumour volume (12 months vs. 7.6 months, respectively, p = 0.04). Radioembolisation also led to a longer time to progression in the liver than chemotherapy alone in the Hendlisz and colleagues86 trial (all progressions considered an event: 5.5 months vs. 2.1 months for the two groups, respectively, p = 0.003). HRs can be seen in Table 34. Hendlisz and colleagues86 report that local progression was documented in four participants (three radioembolisation; one chemotherapy arm) after an unjustified change in the treatment allocated by randomisation (no further details). Censoring these four participants does not change the median time to liver progression (5.6 months vs. 2.1 months, respectively, p = 0.002).

Median time to progression at any site was also shown to be longer in the radioembolisation group compared with the chemotherapy alone group in the Hendlisz and colleagues86 RCT (4.5 months vs. 2.1 months, respectively, p = 0.03). A similar pattern was seen in the study by Van Hazel and colleagues87 (radioembolisation 18.6 months vs. chemotherapy 3.6 months, p < 0.0005). This study also presents the site of the first disease progression. The most common site was the liver (eight participants in each treatment arm). Other sites that showed progression in the radioembolisation group were the liver and lung combined (n = 1), and the lung (n = 1) (one participant in this arm had died without progression). In the chemotherapy-only group, the site of first progression in the other two participants was the liver and peritoneum combined in one, and bone in one other.

In these three studies radioembolisation led to more favourable results in terms of time before progressive disease than chemotherapy alone.

Quality of life

Two included RCTs assessed quality of life as an outcome. 85,87 No numerical outcome data were reported in either of these two studies. Grey and colleagues85 used a 13-point linear analogue self-assessment scale, which was reported as validated. However, no further details of the measure or the validation were provided. The study reports that there were no significant differences between the treatment groups. Sexual interest/ability deteriorated for both treatment groups over the 18-month period during which the protocol treatments were delivered. For all other measures, there were trends towards improvement in quality-of-life scores over the first 18 months for both treatment groups.

Van Hazel and colleagues87 used the 23-point validated Functional Living Index – Cancer questionnaire. 89 The authors state that, at 3 months, changes from baseline patient-rated quality of life were very similar between both arms (p = 0.96). This was also the case for quality of life rated by the physician using the Spitzer index (p = 0.98). 90 The authors suggest that the lack of variation from baseline may have been because most participants were still receiving chemotherapy during this time. In a more recently published conference abstract,88 Van Hazel and colleagues state that at 3 months the radioembolisation group showed a statistically significant improvement in health-related quality of life compared with the chemotherapy-only group (p = 0.03, 95% CI 1.4 to 27.6), but no statistically significant differences thereafter. This appears to be different from the result presented in the full publication; however, there is limited detail reported in either publication and each may, in fact, be presenting different things. As in the earlier publication, no outcome data were reported in the abstract.

Adverse events and toxicity

The total number of complications and toxicities graded 3 or 4 from the included studies can be seen in Table 35. Two RCTs85,87 used the Union International Control Criteria (UICC) to define adverse events and toxicity, and Hendlisz and colleagues86 used the Common Terminology Criteria for Adverse Events (CTCAE, version 3). In the RCT comparing radioembolisation and hepatic artery chemotherapy with hepatic artery chemotherapy, the total number of complications was the same in each of the two treatment groups. 85 In the two RCTs comparing radioembolisation with systemic chemotherapy, one found a greater number of complications in the chemotherapy arm than in the radioembolisation arm,86 and the other found a greater number of complications in the radioembolisation arm than in the chemotherapy arm. 87 In terms of the number of grade 3 or 4 complications, there appears to be no consistent finding across these three studies.

The types of complications and toxicities recorded in the three studies can also be seen in Table 35. These are all those graded 3 or 4. One study86 also reported grade 1 and 2 complications and toxicities, which can be seen in Appendix 4. Overall, the results suggest that radioembolisation is well tolerated, with the occurrence of toxicities and adverse events not appearing to be particularly different from those with hepatic artery chemotherapy or systemic chemotherapy. However, the methodological quality and sample sizes of these included studies would indicate a need for caution in the interpretation of these results.

Summary

Three RCTs85–87 compared radioembolisation with chemotherapy delivered either via the hepatic artery or intravenously. The methodological quality of the RCTs was mixed, in many cases because the methodological details needed to judge different aspects of quality were poorly reported. Two of the RCTs85,87 have an uncertain risk of selection bias which should be considered in the interpretation of these results. The two RCTs86,87 that compared radioembolisation with systemic chemotherapy compared overall survival between the groups. Radioembolisation led to longer median survival in both of these studies; however, the difference between groups was statistically significant in only one study. 87 The third RCT85 which compared radioembolisation with hepatic artery chemotherapy reported median overall survival and estimates of survival at 1, 2, 3 and 5 years but the difference between the groups was not statistically significant.

All three RCTs reported tumour response to treatment but used slightly different criteria for classifying response. Only two RCTs85,86 calculated an estimate of overall response and showed that those in the radioembolisation group had statistically significantly better response rates than their respective comparator groups. The third RCT87 did not report an overall estimate of tumour response rate. Time to progressive disease was reported in different ways in the three included RCTs, which all showed favourable results in the radioembolisation groups when compared with their chemotherapy groups. Quality of life was an outcome in two of the included studies; however, no data were presented. Adverse events and toxicity occurred in all trial arms. In one RCT the adverse events were more numerous in the radioembolisation arm,87 but this pattern was not seen in the other two RCTs. 85,86

Modelling approach

Abramson and colleagues115 report the estimation of the incremental cost of hepatic artery chemoembolisation versus palliative care and of the incremental cost-effectiveness ratios (ICERs) using a range of hypothetical incremental survival estimates over 24 months. There is no description of a model structure and its assumptions or any other indication of modelling the natural progression of the disease. This approach is justified by the lack of evidence on the differential in survival benefit between interventions. However, Abramson and colleagues115 do not report their literature review as systematic. The 11.1-month median survival reported by Stangl and colleagues57 for palliative care justified the use of 12 months of survival benefit for the baseline. 115 Yet details on the search conducted to identify this publication, and the reason for its choice and use of an approximate value is not provided. Results of the case series reported by Abramson and colleagues115 substantiated the speculative range of values used for the incremental survival of hepatic artery chemoembolisation versus palliative care (0–24 months). The rationale for a 2-year model and discounting was not discussed. 115 A lifetime horizon would have been more appropriate;117 however, given the advanced stage of these patients’ disease, 2 years may be close to their lifetime. The assumptions made adopting this type of analysis are not listed and the discussion of them is very limited.

Gazelle and colleagues116 adapted an existing Markov state-transition microsimulation model118 to estimate the relative cost-effectiveness of radiofrequency ablation and surgical resection and a variety of imaging and treatment strategies in people with metastatic colorectal cancer. The modelling approach used is appropriate and the model structure and its assumptions are described in detail. For each strategy, a cohort of 10,000 patients with potential for treatment was simulated, having each individual moving from the ‘alive_treat’ health state to the ‘alive no treat’ or to the ‘dead’ absorbing state, in monthly cycles. Patients eligible for treatment stay in the ‘alive_treat’ health state until they die or become untreatable, that is to say when they present more metastases than the treatment threshold or when they reach the maximum number of interventions for the strategy being analysed. While patients are in the ‘alive_treat’ state, the model tracks metastases progression, diagnosis of the existing metastases (previously undetected), the impact of diagnostic tests and of interventions and their complications.

For each cohort, only one imaging and treatment strategy is modelled by specifying parameters such as the treatment threshold (1–6 metastases), follow-up interval (4, 6 or 12 months), and imaging test sensitivity. For each simulated patient, the model tracks up to 15 individual hepatic metastases, specifying and updating the location, size, rate of growth, detection, and removal/ablation of each metastasis. The initial number, location and size of each metastasis are randomly drawn according to the distributions assigned to each parameter. Over time, metastases grow, may be detected with imaging tests (preoperatively or intraoperative), and may be removed. Each time a patient undergoes a diagnostic imaging test, metastases may be independently detected or missed. Tumour detection is based on test sensitivity, assuming that below a certain (definable) size threshold, all metastases are missed. The authors116 assume that no new liver metastases develop over time.

This microsimulation model also accounted for patients’ quality of life, procedure-related adverse events, and the impact of repeating the procedure. Patients’ baseline survival rates varied according to the percentage of liver volume replaced by tumour (LVRT), but procedure-related morbidity and HRQoL were assumed not to vary according to percentage of LVRT. 116

Gazelle and colleagues116 described and justified the assumptions related to each of these parameters and their sources (Appendix 8, data extraction form). However, their data were derived from a comprehensive review of the literature, which was not reported as systematic and based on expert opinion. Limited detail on parameter derivation is provided, as there is no explanation of the method used to pool estimates from several sources, such as the hazard rates according to the percentage of LVRT. Most estimates used by Gazelle and colleagues116 have come from different data sources and may not be representative of the population of interest. Most input estimates used in the model were applied deterministically, apart from the number of metastases per patient and the size of metastases (which were randomly drawn from a population distribution), and some were subject to sensitivity analysis (further details are provided in Appendix 8b).

Estimation of quality-adjusted life-years

Patients’ HRQoL was accounted for only by Gazelle and colleagues,116 as Abramson and colleagues115 refer to a cost-effectiveness analysis in which health benefits are measured in life-years.

Gazelle and colleagues116 derived the HRQoL data used in their model from a review of the literature (not reported as systematic and described with limited detail) and from expert opinion. 116 According to Gazelle and colleagues,116 the limited data available regarding quality of life in patients with metastatic colorectal cancer suggest that the majority of survival is spent with a normal age-adjusted quality of life and that patients’ quality of life seems to decline rapidly at the end of life. Thus, age- and sex-adjusted population utilities reported by Fryback and colleagues119 [who used a preference measurement method, time trade-off (TTO), to measure the utility for current health of the US general adult population] were applied to survival without liver metastases. Based on evidence from studies120,121 conducted with non-preference-based measures, the quality of life in the month prior to death was assumed to be 60% of that for the control subjects, and a decrement was applied to reflect the impact of the interventions (70% and 95% of the pre-intervention period for surgical resection and ablation respectively). 120,121 None of the values assigned to the health states of patients with liver metastases were estimated through the use of a standardised and validated instrument for HRQoL measurement.

Estimation of costs

Abramson and colleagues115 described in detail the unit costs, resource use, and the sources of the estimates used in their analysis – Abramson and colleagues’ case series data, assumptions, or referenced sources. For instance, the probabilities of re-embolisation and of complications were estimated from the case series reported by Abramson and colleagues. 115 The perspective adopted was that of the third-party payer using 1998 Medicare reimbursement rates as a proxy for inpatient hospitalisation, procedural, and outpatient visit costs, and the Health Care Financing Administration Federal Upper Limit prices in 1998 were used for outpatient drug costs. The resource use considered comprises the initial evaluation of the patient (including specialist visits, CT imaging and blood tests), resources related to the hepatic artery chemoembolisation procedure (including a range of inpatient care accounting for several types of eventual complications), follow-up costs, and iatrogenic symptom control. The costs of disease-related symptom control were assumed to be nearly equal for both arms and therefore not included explicitly.

Gazelle and colleagues116 derived cost data from a non-systematic review of the literature and the methods for their derivation were appropriately described. A societal perspective restricted to productivity and direct medical costs and benefits (including postoperative death costs) was adopted, and the unit costs from the Medicare payment schedule were used for both tests and procedures. The cost of both technical and professional resources for each intervention (liver resection, laparotomy and radiofrequency ablation) and the costs of patient care per year (according to LVRT, derived from Taplin and colleagues122) were included. Concerning radiofrequency ablation, it was assumed that up to three liver metastases could be treated in 1 day and that two treatment sessions would be required for patients undergoing treatment for four to six metastases; thus, the costs were adjusted accordingly, and the cost of repeat radiofrequency ablation was assumed to be the same as the cost of initial therapy. Gazelle and colleagues116 also included morbidity costs related to each intervention, deriving them from daily routine care costs among patients undergoing surgical resection. Complications from radiofrequency ablation were assumed to require 1-day hospitalisation and the resource use involved would be similar to laparotomy-related complications. The cost of preoperative diagnostic CT scans (assumed to be performed on an outpatient basis) and intraoperative ultrasonography were also accounted for.

Primary research

The main characteristics of the included primary studies are briefly presented in Table 38 – further details on each study and their results are reported in Appendix 10.

Study purpose and design

Six of the 14 included studies14,120,121,140–142,144,145,149–151 are intervention studies encompassing the assessment of patients’ HRQoL, and six studies137–139,146–148,151 were aimed specifically at studying the quality of life of patients with liver metastases. One study143 was designed to investigate the clinical outcomes and patient preferences at the end of life. Four RCTs are reported in eight included publications;120,121,140–142,144,149,151 four publications report results from the same RCT. 120,121,140,142 One retrospective cohort study145 and nine prospective cohort studies14,137–139,143,146–148,150 were also included.

Populations

Ten studies14,120,121,137,138,140,142–144,146,147,149–151 involved only participants with colorectal liver metastases. One study139 also enrolled participants with primary liver tumours and benign hepatic lesions, whereas another study148 included patients with liver metastases from other primary tumours (non-small cell lung, pancreas, prostate, and stomach) besides those arising in the colorectum. There was one study on patients with renal liver metastases,141 and another where patients’ primary tumours were not specified. 145

The included studies differed in their participant inclusion criteria and therefore in participants’ prognostic factors. Patients were enrolled in different stages of the disease (resectable,149,151 unresectable,120,121,140,142 unablatable,14 or resected),137 differing as well on the number or extent of metastases. 120,121,140,142,144,149,151

Interventions and comparators

No comparative evidence of the ablative interventions included in the review and the relevant comparators were found. One before-and-after study was found of relevance to the present review – laser ablation. 150 Comparative evidence is available for non-included interventions involving relevant comparators: chemotherapy (hepatic artery chemotherapy and systemic) and BSC,120,121,140–142,144 surgical resection with and without positron emission tomography (PET) imaging,149,151 surgical resection and BSC,138 and surgical resection and/or ablation and BSC following exploratory laparotomy. 14,146,147

Health-related quality-of-life instruments

From the 14 primary studies found on HRQoL of patients with liver metastases, only three report the use of a preference-based measure – the generic instrument European Quality of Life-5 Dimensions (EQ-5D). 146,147,149,151 All 14 studies report the use of a non-preference-based instrument, either generic139,143–145,149,151 or condition-specific,137,138,148,150 or the use of both. 14,120,121,140–142,146,147 No publications involving a condition-specific preference-based measure were found.

The following generic non-preference-based instruments were used among the studies included: EQ visual analogue scale (VAS),14,141,146,147,149,151 Short Form questionnaire-12 items (SF-12),139 Karnofsky Performance Scale index,121,138,142,145 Sickness Impact Profile (SIP),121,140,142 Hospital Anxiety and Depression Scale (HADS),120,121,140,142 Rand 36-item health status profile (RAND-36), Memorial Symptom Assessment Scale, Medical Outcomes Study Sexual Functioning Scale,144 functional scale by Katz, quality of life (QoL) Excellent–Poor scale and Profile of Mood States (POMS). 143

Nine of the 14 studies using non-preference-based measures used condition-specific measures. These were the European Organisation for Research and Treatment of Cancer quality-of-life questionnaire (EORTC QLQ-C30),14,138,146–148,150 its liver metastases-specific questionnaire module EORTC QLQ-LMC21,138 the Functional Assessment of Cancer Therapy – General scale (FACT-G) and its kidney-specific supplement the Functional Assessment of Cancer Therapy – Kidney Symptom Index – Disease-Related Symptoms Subscale (FKSI-DRS),141 the City of Hope QoL Scale/Cancer Patient Survey,137 and the Rotterdam Symptom Checklist (RSC). 120,121,140,142

One study using both condition-specific (EORTC QLQ-C30) and generic (EQ-VAS) non-preference-based measures reports similar profiles of quality-of-life variation over time for both measures. 153 Although some evidence suggests that both preference- and non-preference-based instruments capture the variation of people’s quality of life over time,146 results for patients with unresectable liver metastases suggest that non-preference-based (generic and condition-specific) and preference-based measures are sensitive to different aspects of quality of life over time. 147 For resectable/ablatable patients, the condition-specific measure appeared to be more sensitive to variations in patients’ quality of life. 147 Despite the slight discrepancies in results from studies with few participants, the EQ-5D seems to capture appropriately the quality of life of patients with liver metastases, as its pattern of scores over time was similar to those obtained with the non-preference-based EQ-5D VAS (generic) and EORTC QLQ-C30 (cancer-specific) measures. 146,147

Preference-based measures

Three studies129,146,149,151 reported utility values for people with liver metastases. Each individual study investigated the HRQoL of fewer than 100 patients with colorectal liver metastases. Two studies149,151 report results from one RCT151 to assess the added value of PET imaging that subsequently led to a prospective cohort study of the total population,149 whereas the other two publications146,147 are prospective cohort studies aimed specifically at the study of HRQoL in patients with liver metastases. Table 39 shows a summary of the utility values reported in the primary research included in this review.

Wiering and colleagues149,151 provide EQ-5D utilities for resectable patients at baseline and after surgical resection over 3 years in different health states: disease-free, non-curative, and recurrence (treated with surgical resection or chemotherapy). Patients in the different health states had baseline scores crudely close to 0.8 and a clear decrease 3 weeks after surgical resection,151 suggesting that patients with liver metastases have similar HRQoL to that of the general population at that age. The Health Survey for England (1996) reports the mean (± standard error) utilities for people living in private households in England as 0.79 (± 0.006) for 55–64-year-olds and 0.78 (± 0.006) for 65- to 74-year-olds. 154

Two other studies146,147 provide EQ-5D index scores over time for patients undergoing exploratory laparotomy. The design of these studies146,147 does not allow the differentiation between surgical resection and ablation in their impact on patients’ quality of life. Results for patients with unresectable liver metastases (n = 20) suggest that non-preference-based (generic and condition-specific) and preference-based measures are sensitive to different aspects of quality of life over time, as non-preference-based scores presented deterioration of HRQoL at 6 months after enrolment, whereas EQ-5D utilities improved slightly 2 weeks after enrolment but became comparable with baseline scores at 3 and 6 months. 147 These results should be considered with caution given that this is a prospective study of a small cohort.

Estimates for the unresectable/unablatable group reported by Langenhoff and colleagues147 are similar to those obtained by Krabbe and colleagues. 146

Effect of liver metastases on patients’ quality of life

Some evidence from generic non-preference-based measures suggests that liver metastases do not have much impact on patients’ overall quality of life. 121,137,142,143 This may be a result of the lack of sensitivity of the instruments to the health domains affected by liver metastases. Condition-specific measure EORTC QLQ-LMC21 symptom scales and single items indicate the main symptoms affecting patients with liver metastases. There is evidence that patients with liver metastases on palliative care report significant deterioration over time of symptoms, such as taste problems, dry mouth, sore mouth, and peripheral neuropathy. 138 Patients with liver metastases show variation in quality of life according to the severity of the disease (as different baseline EQ-5D and EORTC scores are shown in studies reporting results by resectable/ablatable/unresectable subgroups),14,138,146,147 and to the recurrence of liver metastases. 151

No significant difference in functional improvement was shown between patients with ≥ 90% or < 90% total liver tumour volume resected. 145 Contrastingly, Earlam and colleagues121 showed a positive correlation between percentage of hepatic replacement and RCS, SIP and HADS scores. There is also evidence that the quality of life of patients with liver metastases is influenced by the percentage of hepatic replacement,121 and by the percentage of tumour removed (non-curative surgery). 151 These results suggest that Karnofsky Performance Scale145 is not sensitive to variations on quality of life of patients with liver metastases according to hepatic volume replaced by tumour.

The quality of life of patients on symptom control seems to deteriorate late in the disease course (a significant decrement of quality of life is only seen 5 months to 1 month before death) and patients receiving symptom control spent approximately 24% of their survival with normal RSC, SIP and HAD scores. 120,121 Unablatable patients had no clear variation of quality of life (EQ-VAS and EORTC QLQ-C30 scores) at 2 weeks and 3 months after enrolment, showing deterioration only at 6 months. However, improvement of the EQ-5D index scores was seen 2 weeks after enrolment and these scores were comparable with baseline at 3 and 6 months. 147 Unresectable patients had worse EORTC QLQ-LMC21 symptoms scores at 3 months (taste problems, dry mouth, sore mouth and peripheral neuropathy). 138

The quality of life of patients with liver metastases appears to vary according to the stage of the disease and prognostic factors, such as number and size of liver metastases (surrogate indicators: hepatic volume replaced by tumour, eligibility for surgical resection/ablation), and seems to deteriorate slowly over time, weakening severely at the end of life.

Secondary research

The three reviews which met the inclusion criteria were economic evaluations using utility values to calculate QALYs (Table 40). 116,118,136 Two of these were conducted by Gazelle and colleagues,116,118 who performed a cost–utility analysis to estimate the relative cost-effectiveness of radiofrequency ablation and surgical resection (addressed also in Systematic review of existing cost-effectiveness evidence) through the adaptation of the model they had previously used to study the cost-effectiveness of surgical resection compared with no treatment. 118 Gazelle and colleagues used the same rationale and HRQoL data sources for the estimation of QALYs in both economic evaluations, based on published evidence119–121 and expert opinion. From the Earlam and colleagues121 and Allen-Mersh and colleagues120 studies (both included in the current review), Gazelle and colleagues116 inferred that the majority of survival of patients with colorectal cancer liver metastases is spent with a normal age-adjusted quality of life. Thus, age- and sex-adjusted population utilities119 were applied to survival without liver metastases (following surgical resection/ablation). A decrement was applied to reflect the impact of the interventions (70%116,118 and 95%116 of the pre-intervention period for surgical resection and ablation, respectively). The quality of life of patients with liver metastases in the month prior to death was also assumed to be 60% of that for the control subjects. No sensitivity analyses were conducted on these estimates.

Karuna and colleagues136 conducted a cost–utility analysis of laparoscopy versus laparotomy for initial surgical assessment and treatment of potentially resectable colorectal liver metastases. The utility weights used for the estimation of QALYs were derived from the HRQoL study by Langenhoff and colleagues147 (included in the current review) and a HRQoL study in hepatocellular carcinoma patients155 (a different population, hence not included in this review). Patients with immediate post-operative complications were assigned a small decrement (0.02) in utility to account for the additional loss of quality of life. Patients’ quality of life was also assumed to be constant after 6 months post intervention, which was supported by recent evidence that showed that quality of life is stable from 6 months post intervention until immediately prior to death. 119,147,155

Implications for future cost-effectiveness modelling

Modelling options are restricted by the little evidence available on HRQoL of patients with liver metastases. The evidence available suggests near-normal quality-of-life levels at baseline, substantial decrement due to the intervention followed by a quick recovery to baseline values138,146,147,149,151,153 and serious deterioration of quality of life at the end of life,120,121,140,142 supporting the rationale taken on in previously published economic evaluations. 116,118,136 Evidence also suggests that patients’ eligibility to the interventions varies according to disease severity (resectable, ablatable, inoperable). 138,146,147 The quality of life of patients treated by surgical resection or ablation does not deteriorate as much as that for patients found to be unresectable/unablatable after exploratory laparotomy, and this latter group takes longer to regain its baseline quality of life. 147 Ablatable patients presented better scores than resectable/ablatable ones over 6 months, while those found to be unresectable at laparotomy reported worse scores over 6 months for EQ-VAS, EORTC QLQ-C30, and EQ-5D index. 146

Moreover, statistically significant differences in quality of life over time were found between groups in different health states (disease free, non-curative, and recurrence) indicating different quality of life for progression-free survival and post-progression survival of resected patients and the impact of non-curative interventions. 151 Also, the RSC physical score was shown to be a predictor of survival for patients on symptom control,121 and patients’ RAND-36 physical functioning was shown to vary over time as a function of survival. 144

Implications for the independent survival model

In order to capture the most significant variations of HRQoL of patients with liver metastases over time, the independent survival-based model (further described in Independent economic evaluation) includes a number of health states. The impact of the interventions is captured by the resultant life expectancy in each of the different utility-weighted health states.

The comparisons being analysed in the section of the report headed Independent economic evaluation (regarding radioembolisation relative to hepatic arterial infusion, hepatic arterial infusion alone, microwave ablation, radiofrequency ablation, and surgical resection) involve different subgroups of patients. The comparison of radioembolisation associated to hepatic arterial infusion with hepatic arterial infusion alone investigated unablatable patients (non-eligible to any sort of surgical or local ablative intervention);85 hence, the utility values reported for the inoperable group by Langenhoff and colleagues147 were assumed to be generalisable to these patients. Both Shibata and colleagues72 and Kim and colleagues73 enrolled resectable patients to compare an ablative intervention with surgical resection. Although patients’ inclusion criteria (such as the maximum number of metastases) differ among studies, Wiering and colleagues’151 utility estimates seem generalisable to the surgical resection arms of the clinical effectiveness studies. 72,73 Similarly, utilities from Krabbe and colleagues’146 ablatable group were used as estimates for the impact of the ablative therapies on quality of life and for ablated patients with stable disease.

The following assumptions underlie the choice of the estimates presented in Table 41:

• Interventions’ impact measured 2 weeks after intervention is assumed to remain for a whole month as it is applied to the first cycle of the model.

• Given the lack of specific estimates, all ablative therapies are assumed to have the same impact on patients’ quality of life.

• Ablated patients are assumed to reach stable disease at 6 months.

• Utility weight for resected patients with progressive disease assumed to be the same for ablated patients in this health state, given the lack of particular evidence. This is thought to be a conservative approach, as ablatable patients presented better scores than resectable/ablatable ones over 6 months for EQ-VAS, EORTC QLQ-C30, and EQ-5D index. 146

• The same utility weight was assigned to all patients with terminal disease irrespective of their initial disease severity and the interventions they had undertaken. This is the estimate from the final measurement by Wiering and colleagues151 for their recurrence group with most advanced disease.

Model type and rationale for model structure

Figure 4 presents a schematic of the basic survival model which, in its simplest form, contains three states – stable disease (i.e. patients’ state at entry to the trial), progressive disease and death. 156,157 Movements between these states are usually permitted only in the progressive direction. This approach has been adopted to model the cost-effectiveness of ablative therapies for liver metastases.

FIGURE 4.

Schematic of the survival model adopted for the cost-effectiveness model. OS, mean overall survival; OS-TTP, mean survival duration with progressive disease; TTP, mean time to progression.

An advantage of this approach is that the model is based on relevant final outcomes (survival weighted by quality of life) rather than using surrogate measures, such as tumour response or complete/partial resection for which additional evidence would be required to establish the link to final outcomes. Interpretation of this basic model in the context of metastatic disease is slightly more complex than in the usual application, where therapy is directed at the primary tumour. In this case patients’ primary tumour is assumed to be stable and the intention of treatment is to remove the metastases (in the case of treatment with curative intent) or to reduce the size of the metastases, improving quality of life and offering some increase in life expectancy, in the case of treatment with palliative intent. Therefore, the stable disease state may include quite different patients – those whose metastases have been completely removed and those with remaining metastases. Quality of life for this group of patients is weighted to take account of those metastases that are removed (equivalent to a complete response on standard chemotherapy outcome scales) and those with remaining metastases (see later in this section). Including both groups of patients in a single ‘stable’ health state in the survival model, rather than adopting a state transition model (explicitly modelling response/removal of metastases), is a pragmatic decision based on the fact that studies have not reported survival, duration of response or time to disease progression for these patient groups separately.

Patients enter the model with a stable primary tumour, with liver metastases and receive treatment with an ablative therapy (with curative or palliative intent depending on the patient population in the included study) or the relevant comparator treatment. Patients may experience disease progression or may die without experiencing documented disease progression.

The model was developed in Microsoft Excel (Microsoft Corporation, Redmond, WA, USA). Two versions of the model were developed independently, using the same input parameters, and tested against each other to ensure internal consistency. It is fully probabilistic to take into account parameter imprecision. In addition, deterministic sensitivity analysis was used to explore different scenarios and assumptions in the model.

The base-case analysis compares the mean overall survival with ablative therapy (OS¯AT) with the mean overall survival for comparator (OS¯C). he estimate of life-years gained with the ablative therapy, in the base case, was calculated as:

To estimate the QALY gain associated with the ablative therapy (QALYG_AT), response-specific utilities [U_S for stable (which may be a weighted sum of the utility in patients whose metastases are removed and those successfully treated but with residual metastases) and U_P for disease progression, respectively], derived in our review of quality-of-life studies (see Systematic review of health-related quality-of-life evidence), were applied to the mean overall survival estimates, taking into account survival before disease progression (mean time to progression, TTP¯AT) and post-progression survival (OS¯AT−TTP¯AT). The quality-adjusted life expectancy gain was therefore calculated as:

Baseline cohort

The population in the base-case analysis are adult patients with liver metastases. Patients receiving ablative therapies, other non-invasive therapies or comparator treatments are assumed to have a stable primary tumour and may be undergoing treatment of curative or of palliative intent.

Perspective and time horizon

The perspective of the cost-effectiveness analysis is that of the NHS and PSS. A lifetime time horizon has been adopted for the model.

Discounting

Both costs and outcomes were discounted using a 3.5% discounting rate, as currently recommended by the UK Treasury for public sector appraisal. 158

Assessment of uncertainty

The purpose of this analysis is to test the robustness of the cost-effectiveness results to variation in structural assumptions and parameter inputs. Deterministic analyses were used to address particular areas of uncertainty in the model. The uncertainties around the probability, resource use and cost estimates that were expected, a priori, to have a disproportionate impact on the study results were investigated by applying ranges around the point estimates used in the base-case analysis. Scenario analysis was used to address the uncertainty associated with the choice of data source adopted for parameter values in the base case and some aspects of the chosen structure of the model.

Parameter uncertainty was addressed using probabilistic sensitivity analysis (PSA). Probability distributions were assigned to describe uncertainty around the point estimates used in the base-case analysis. Variables included in the PSA, the sampling distribution and the parameterisation of the sampling distribution are reported in Appendix 11. Each PSA is run for 1000 iterations.

Clinical effectiveness

Statistical models were fitted to data derived from survival plots reported in the included studies. Three survival functions (exponential, Weibull and log-logistic) were fitted in each case and the function offering the best fit was selected for the survival model. The three survival functions have different assumptions regarding the hazard function: the exponential function assumes constant hazards over time; the Weibull function assumes that hazard either increases or decreases monotonically with time; and the log-logistic function has a hazard function with a single peak, thus the hazard may increase until time t and decrease thereafter.

The exponential model has a very simple form and can be derived easily with minimal reported data (e.g. from median survival),159 which may explain its popularity in many health economic models. However, the constant hazard property may not be appropriate in many clinical settings. In contrast, the log-logistic function, while having the flexibility to capture monotonically declining rates as well as those that initially increase and subsequently decline, can give rise to implausibly high maximum survival durations. As a rule, the simplest form that is consistent with the observed data will be selected, which will, in many cases, be the Weibull function.

Overall survival – microwave ablation compared with surgical resection

Overall survival curves for patients undergoing surgical resection and microwave ablation, reported by Shibata and colleagues,72 were scanned using Engauge software (Engauge Digitizer: http://digitizer.sourceforge.net) and imported into Microsoft Excel. At the end of follow-up all patients with complete follow-up in the trial had died [12 in the surgical resection arm (seven with hepatic failure) and nine in the microwave ablation arm (six with hepatic failure)], with four patients in the surgical resection arm and five in the microwave ablation arm censored at between 10 and 30 months of follow-up.

The extrapolated overall survival curves for the surgical resection arm, using different survival functions, are given in Figure 7 along with Kaplan–Meier survival estimates from the trial report. The Weibull and log-logistic survival functions appear to give the closest fit to the overall survival curves.

FIGURE 7.

Kaplan–Meier survival estimates from the Shibata and colleagues72 trial showing Weibull and alternative model fit.

The survival model was fitted to the survival plots for both interventions simultaneously, using linear regression, with surgical resection as the base intervention and estimating a HR for microwave ablation, relative to surgical resection – full details on the method for transforming data for use in the linear regression and for estimating the survival models are reported in Appendix 12.

Table 42 reports a comparison of observed survival at 1, 2 and 3 years against model predictions.

TABLE 42 - Comparison of observed survival at 1, 2 and 3 years against model predictions

MWA, microwave ablation.

Year	Trial report		Weibull model		Exponential model		Log-logistic model
	MWA	Surgical resection	MWA	Surgical resection	MWA	Surgical resection	MWA	Surgical resection
1	71%	69%	82%	83%	48%	40%	80%	78%
2	57%	56%	43%	47%	23%	16%	34%	31%
3	14%	23%	14%	18%	11%	8%	13%	12%

Data in Table 42 suggest that, while the exponential survival function is a particularly poor fit for the observed data, none of the survival functions provides good predictions of survival for the reported time periods. Both the Weibull and log-logistic functions overestimate early survival (up to 1 year) and underestimate later survival (from 2 years). The exponential and log-logistic functions estimate lower survival with surgical resection at each reported time point, while the observed data has a higher survival with surgical resection at 3 years.

Mean survival reported by Shibata and colleagues72 and from each of the modelled survival functions (extrapolated to a maximum duration of 49 months for comparison with the area under the reported Kaplan–Meier curve) are reported in Table 43.

TABLE 43 - Mean overall survival from Kaplan–Meier and modelled survival functions

MWA, microwave ablation.

Treatment arm	Mean overall survival (months)
	Trial report	Weibull	Exponential	Log-logistic
Surgical resection	24.7	24.1	12.7	21.1
MWA	24.8	22.9	15.4	21.7

Mean overall survival for the surgical resection arm estimated using the Weibull survival function is similar to that estimated from the Kaplan–Meier curve and the value reported in the trial publication (24.7 estimated from the Kaplan–Meier curve, 24.1 using the Weibull survival function and 25 months reported in the trial publication). However, while mean survival estimated from the Kaplan–Meier curve for microwave ablation is the same as for surgical resection (and in the trial publication is higher, at 27 months), mean overall survival for microwave ablation, estimated using the Weibull survival function (22.9 months), is lower than for surgical resection. The model using the exponential function appears to substantially underestimate overall survival for both surgical resection and microwave ablation (with lower estimated survival for surgical resection compared with microwave ablation). The log-logistic survival function also underestimates overall survival for both surgical resection and microwave ablation (with lower estimated survival for surgical resection than for microwave ablation), though the differences compared with the reported trial values are not as great.

The Weibull model for the base-case analysis was adopted and the alternative survival functions in structural sensitivity analyses were included.

Radiofrequency ablation compared with surgical resection for solitary metastases < 3 cm

The extrapolated overall survival curves, for surgical resection, are given in Figure 8 along with Kaplan–Meier survival estimates from the trial report. The Weibull and log-logistic survival functions appear to fit the overall survival curves reasonably well, although all of the modelled functions give lower predictions than the Kaplan–Meier curves for survival over 5 years. The Weibull and log-logistic functions appear to overestimate median survival [trial report = 59.6 months (this is not reported in the study, but has been estimated from the Kaplan–Meier curve), Weibull = 52.9 months, log-logistic = 54.5 months] while the exponential provides a closer estimate for median overall survival (61.2 months).

FIGURE 8.

Kaplan–Meier survival estimates for resected patients with solitary metastases (< 3 cm) from the Kim and colleagues73 study showing alternative model fit.

Table 45 reports a comparison of observed survival at 1, 2, 3 and 5 years against model predictions.

Data in Table 45 suggest that the exponential survival function is a poor fit to the observed data over the first 24 months, while the Weibull and log-logistic functions both provide a reasonable fit for early survival (up to 3 years). Both functions seem to underestimate later survival with surgical resection and RFA. In contrast the exponential function appears to give very close predictions for 3- and 5-year survival for surgical resection.

Mean survivals derived from the study by Kim and colleagues73 and from each of the modelled survival functions (extrapolated to a maximum duration of 72 months for comparison with the area under the reported Kaplan–Meier curve) are reported in Table 46.

The exponential and Weibull give close predictions for mean overall survival, although the Weibull function indicates no difference in survival between surgical resection and RFA, while the values read from the Kaplan–Meier curve suggest that survival is slightly higher with surgical resection. The log-logistic function slightly overestimates overall survival compared with the values read from the Kaplan–Meier curve and suggests a slightly greater survival difference.

The Weibull model was adopted for the base-case analysis and the alternative survival functions in structural sensitivity analyses were included.

Radiofrequency ablation compared with surgical resection for solitary metastases ≥ 3 cm

The extrapolated overall survival curves, for surgical resection, using different survival functions, are given in Figure 9 along with Kaplan–Meier survival estimates from the trial report. The Weibull and log-logistic survival functions appear to fit the overall survival curves reasonably well, although all of the modelled functions give lower predictions than the Kaplan–Meier curves for survival over 5 years (48% from the Kaplan–Meier curve vs. 43% for the exponential, 42% for Weibull and 43% for the log-logistic model). Each of the modelled functions appears to overestimate median survival [trial report = 45.6 months (this is not reported in the study, but has been estimated from the Kaplan–Meier curve), exponential = 49.5 months, Weibull = 50.6 months, log-logistic = 50.2 months].

FIGURE 9.

Kaplan–Meier survival estimates for resected patients with solitary metastases (≥ 3 cm) from the Kim and colleagues73 study showing alternative model fit.

Table 47 reports a comparison of observed survival at 1, 2, 3 and 5 years against model predictions.

Data in Table 47 suggest that the exponential survival function underestimates early (up to 1 year) and late (5 years) survival, while the Weibull and log-logistic functions both provide a reasonable fit for early survival (up to 2 years). Both functions seem to overestimate survival at 3 years and to underestimate survival at 5 years.

Mean survival derived from the study by Kim and colleagues73 and from each of the modelled survival functions (extrapolated to a maximum duration of 72 months for comparison with the area under the reported Kaplan Meier curve) is reported in Table 48.

Mean overall survival for surgical resection estimated using the Weibull and log-logistic survival functions is similar to that reported from the trial (47.2 months for the modelled functions compared with 47.4 months from the Kaplan–Meier curve). However, both functions estimate higher mean survival with radiofrequency ablation (34.2 and 34.5 months for Weibull and log-logistic functions, respectively) compared with the area under the Kaplan–Meier curve (32.9 months). The model using the exponential function appears to provide a particularly poor fit for both surgical resection and radiofrequency ablation.

The Weibull model was adopted for the base-case analysis and the alternative survival functions in structural sensitivity analyses were included.

Overall survival – radioembolisation plus hepatic artery chemotherapy compared with hepatic artery chemotherapy alone

Overall survival curves reported by Grey and colleagues85 were scanned using Engauge software and imported into Microsoft Excel. At the end of follow-up four patients included in the trial were still alive (three in the radioembolisation plus hepatic artery chemotherapy arm and one in the control arm). The final portions of the survival curves were extrapolated using a regression analysis.

The extrapolated overall survival curves, for radioembolisation plus hepatic artery chemotherapy, using different survival functions, are given in Figure 10 along with Kaplan–Meier survival estimates from the trial report. The Weibull and log-logistic survival functions appear to fit the overall survival curves reasonably well, with the log-logistic survival function providing the closest fit for median survival (trial report = 17 months, log-logistic = 18 months, Weibull = 20 months, exponential = 15.6 months).

FIGURE 10.

Kaplan–Meier overall survival estimates for radioembolisation plus hepatic artery chemotherapy from the Grey and colleagues85 trial showing alternative model fits. HAC, hepatic artery chemotherapy.

Table 49 reports a comparison of observed survival at 1, 2, 3 and 5 years against model predictions.

TABLE 49 - Comparison of observed survival at 1, 2, 3 and 5 years against model predictions

HAC, hepatic artery chemotherapy; RE, radioembolisation.

Year	Trial report		Weibull model		Exponential model		Log-logistic model
	HAC	RE + HAC	HAC	RE + HAC	HAC	RE + HAC	HAC	RE + HAC
1	68%	72%	62%	72%	47%	59%	61%	71%
2	29%	39%	29%	41%	22%	34%	34%	36%
3	6%	17%	11%	21%	10%	20%	13%	19%
5	0%	4%	1%	4%	2%	7%	5%	7%

Data in Table 49 suggest that the exponential survival function is a poor fit to the observed data, while the Weibull and log-logistic functions both provide a reasonable fit for early survival (up to 2 years). Both functions seem to overestimate survival with hepatic artery chemotherapy at 3 years, with the log-logistic also overestimating survival for radioembolisation plus hepatic artery chemotherapy and hepatic artery chemotherapy alone. In contrast, the Weibull function appears to give reasonable predictions of 5-year survival for both treatment arms.

Mean survival reported by Grey and colleagues85 and from each of the modelled survival functions (extrapolated to a maximum duration of 57 months for comparison with the area under the reported Kaplan–Meier curve) is reported in Table 50.

TABLE 50 - Mean overall survival from Kaplan–Meier and modelled survival functions for radioembolisation plus hepatic artery chemotherapy and hepatic artery chemotherapy alone

HAC, hepatic artery chemotherapy; RE, radioembolisation.

Treatment arm	Mean overall survival (months)
	Trial report	Weibull	Exponential	Log-logistic
RE + HAC	23.5	23.3	20.7	22.7
HAC	18.4	18.6	15.5	19.0
Difference	5.1	4.7	5.2	3.7

Mean overall survival for radioembolisation plus hepatic artery chemotherapy and for hepatic artery chemotherapy alone estimated using the Weibull survival function is very similar to that reported from the trial (23.3 compared with 23.5 months and 18.6 compared 18.4 months, for radioembolisation plus hepatic artery chemotherapy and for hepatic artery chemotherapy alone, respectively). The model using the exponential function appears to underestimate overall survival for both radioembolisation plus hepatic artery chemotherapy and for hepatic artery chemotherapy alone, and to overestimate the survival difference when compared with the trial report. In contrast, while the log-logistic survival function gives the closest fit for median survival it appears to underestimate mean survival for radioembolisation plus hepatic artery chemotherapy while overestimating for hepatic artery chemotherapy alone, hence underestimating the survival difference compared with the trial report.

The Weibull model was adopted for the base-case analysis and the alternative survival functions in structural sensitivity analyses were included.

Progression-free survival – microwave ablation compared with surgical resection

The trial report by Shibata and colleagues72 does not report Kaplan–Meier curves for progression-free survival, or the proportion of patients remaining progression free at given time points. The only reported information on time to disease progression reported in the trial publication is the mean disease-free interval – no measures of variability (SDs or CIs) are reported. These mean values (Table 51) are used in the model as estimates of progression-free survival – in the absence of any further indication an exponential form for the survival function was assumed.

Progression-free survival – radiofrequency ablation compared with surgical resection

Radiofrequency ablation compared with surgical resection for solitary metastases < 3 cm

The extrapolated progression-free survival functions for surgical resection are given in Figure 11 along with Kaplan–Meier estimates from the study publication. The Weibull and log-logistic survival functions appear to fit reasonably well for up to 12 months, but overestimate survival from 12 to 30 months. All the parametric functions appear to overestimate median survival with the exponential function (at approximately 30 months) giving the greatest discrepancy with the Kaplan–Meier estimate.

FIGURE 11.

Kaplan–Meier progression-free survival estimates for resected patients with solitary liver metastases (< 3 cm) from the Kim and colleagues73 study showing alternative model fit.

Mean times to progression from each of the modelled survival functions (truncated at 72 months for comparison with the area under the reported Kaplan–Meier curve) are reported in Table 52.

TABLE 52 - Mean progression-free survival from Kaplan–Meier and modelled survival functions for patients with solitary liver metastases (< 3 cm)

RFA, radiofrequency ablation.

Treatment arm	Mean progression-free survival (months)
	Study report	Exponential	Weibull	Log-logistic
Surgical resection	30.0	34.3	27.8	29.2
RFA	31.6	34.4	29.0	30.5
Difference	1.6	0.1	1.2	1.3

Both the Weibull and log-logistic functions appear to underestimate mean progression-free survival, and to slightly underestimate the difference between surgical resection and radiofrequency ablation. The exponential function shows the greatest discrepancy with the values observed in the study.

The Weibull model was adopted for the base-case analysis and the alternative survival functions in structural sensitivity analyses were included.

Radiofrequency ablation compared with surgical resection for solitary metastases ≥ 3 cm

The extrapolated progression-free survival curves, for surgical resection, using different survival functions, are given in Figure 12 along with Kaplan–Meier survival estimates from the trial report. The Weibull and log-logistic survival functions appear to fit the progression-free survival curves reasonably well, although all the modelled functions give lower predictions than the Kaplan–Meier curves for survival over 5 years. Each of the modelled functions appears to overestimate progression-free survival [trial report = 24.9 months (this is not reported in the study, but has been estimated from the Kaplan–Meier curve), exponential = 37.1 months, Weibull = 32 months, log-logistic = 30 months].

FIGURE 12.

Kaplan–Meier progression-free survival estimates for surgical resection in patients with solitary metastases (≥  3 cm) from the Kim and colleagues73 study showing alternative model fit.

Mean progression-free survival derived from the study by Kim and colleagues73 and from each of the modelled survival functions (extrapolated to a maximum duration of 72 months for comparison with the area under the reported Kaplan–Meier curve) is reported in Table 53.

TABLE 53 - Mean progression-free survival from Kaplan–Meier and modelled survival functions for patients with solitary metastases (≥ 3 cm)

RFA, radiofrequency ablation.

Treatment arm	Mean overall survival (months)
	Study report	Exponential	Weibull	Log-logistic
Surgical resection	35.5	39.1	37.0	36.7
RFA	20.1	39.3	17.1	18.5
Difference	15.1	0.2	19.9	18.3

The Weibull model was adopted for the base-case analysis and the alternative survival functions in structural sensitivity analyses were included.

Progression-free survival – radioembolisation plus hepatic artery chemotherapy compared with hepatic artery chemotherapy alone

Kaplan–Meier estimates for time to disease progression (by tumour area and by tumour volume) reported by Grey and colleagues85 were scanned using Engauge software and imported into Microsoft Excel. The extrapolated survival functions for time to disease progression for radioembolisation plus hepatic artery chemotherapy are given in Figure 13 along with Kaplan–Meier survival estimates from the trial report. The Weibull and log-logistic survival functions appear to fit reasonably well, with both survival functions providing close fits for median time to disease progression (trial report = 15.9 months, Weibull = 15.7 months, log-logistic = 15.4 months, exponential = 14.9 months).

FIGURE 13.

Kaplan–Meier time to progression estimates from the Grey and colleagues85 trial showing Weibull and alternative model fit. HAC, hepatic artery chemotherapy; RE, radioembolisation.

Mean times to progression from each of the modelled survival functions (truncated at 24 months for comparison to the area under the reported Kaplan–Meier curve) are reported in Table 54.

Mean progression-free survival for radioembolisation plus hepatic artery chemotherapy and for hepatic artery chemotherapy alone estimated using the Weibull survival function is the same as the value estimated from the Kaplan–Meier curves reported from the trial (15.2 months compared with 15.2 months and 9.8 months compared with 9.8 months, for radioembolisation plus hepatic artery chemotherapy and for hepatic artery chemotherapy alone, respectively), with similar estimates of the difference. The model using the exponential function appears to underestimate progression-free survival for radioembolisation plus hepatic artery chemotherapy but overestimate for hepatic artery chemotherapy alone, leading to an underestimate of the difference in progression-free survival when compared with the trial report and the Weibull survival model. In contrast, while the log-logistic survival function appears to give a similar fit to the Weibull function (see Figure 13), it underestimates progression-free survival for both radioembolisation plus hepatic artery chemotherapy and for hepatic artery chemotherapy alone, and appears to overestimate the survival difference compared with the trial report and the Weibull survival model.

The Weibull model was adopted for the base-case analysis and the alternative survival functions in structural sensitivity analyses were included.

Adverse events – microwave ablation compared with surgical resection

Adverse events for patients in the study by Shibata and colleagues72 are reported in Table 13 in the clinical effectiveness section of this report (see Chapter 4, Microwave ablation).

Adverse events – radiofrequency ablation compared with surgical resection

The study by Kim and colleagues73 reports that there was no treatment-related mortality in the radiofrequency ablation or surgical resection groups, but that treatment-related morbidity was statistically significantly higher in the surgical resection group [11/177 (6.2%) of the radiofrequency ablation group compared with 59/278 (21.2%) of the surgical resection group, p < 0.001]. The number and proportion of patients in each group experiencing treatment-related morbidity, by cause, is reported in Table 55.

Adverse events – radioembolisation plus hepatic artery chemotherapy compared with hepatic artery chemotherapy alone

Adverse events for patients in the study by Grey and colleagues85 are reported in Table 35 in the clinical effectiveness section of this report (see Chapter 4, Radioembolisation).

Resource use and cost

The groups of health care costs included in the base-case health economic model are treatment costs [including cost of interventions, costs of intervention administration, on-treatment monitoring and costs of post-discharge follow-up (where relevant)], post-treatment monitoring and palliative care costs.

Results of independent economic analysis

Probabilistic analysis

In a PSA, where the parameters of the survival models (both overall and progression-free survival), health state utility values, treatment cost, cost of outpatient attendances and patient monitoring, as well as costs of managing adverse events and palliative care were sampled probabilistically, microwave ablation is associated with reduced costs in 56% of simulations (with a range from –£10,866 to £8606) and lower QALYs in 68% of simulations (with a range of –0.80 to 0.54) when compared with surgical resection (Figure 14 shows this and also shows the 95% confidence ellipse).

FIGURE 14.

Cost-effectiveness plane – incremental cost and incremental QALYs for microwave ablation compared with surgical resection.

The probabilistic analysis generated cost and QALY estimates for each intervention that were greater than those for the base-case analysis (see Table 71 for the base-case analysis). However, the mean incremental cost and mean QALY gain were similar (–£327 and –0.09 QALYs compared with –£377 and –0.07 QALYs for the deterministic base case and PSA respectively). Table 73 reports the mean costs and outcomes from the probabilistic analysis (including the 2.5th and 97.5th percentiles to give an indication of the range of the simulated values) and the ICER for microwave ablation compared with surgical resection, based on the mean values generated in the probabilistic analysis.

TABLE 73 - Costs and outcomes from probabilistic analysis for microwave ablation and surgical resection

MWA, microwave ablation.

Procedure	Discounted costs (£)			Discounted QALYs			ICER (£)
	Mean	2.5th percentile	97.5th percentile	Mean	2.5th percentile	97.5th percentile
Surgical resection	21,578	13,298	35,722	1.44	0.59	2.61
MWA	21,160	13,091	35,529	1.37	0.61	2.55	5428

In addition to graphing the incremental cost and incremental QALYs for microwave ablation, a cost-effectiveness acceptability curve was derived, representing the proportion of simulations where microwave ablation is cost-effective for a range of willingness-to-pay thresholds, up to £100,000 (Figure 15). In this analysis microwave ablation had a probability of being cost-effective of 31% at a willingness-to-pay threshold of £20,000 per QALY and 30% at a willingness-to-pay threshold of £30,000 per QALY.

FIGURE 15.

Cost-effectiveness acceptability curve for microwave ablation.

Radiofrequency ablation compared with surgical resection

This section reports cost-effectiveness results for a cohort of patients with resectable liver metastases (small or large solitary liver metastases) undergoing either radiofrequency ablation or surgical resection, based on data reported by Kim and colleagues. 73 The analyses are conducted separately for each patient population, those with:

• solitary metastases < 3 cm

• solitary metastases ≥ 3 cm.

Discounted costs (identifying the contribution of procedure costs, surgical follow-up, monitoring prior to disease progression and palliative care) are presented alongside the life expectancy and quality-adjusted life expectancy for patients in the cohort. The results are presented as incremental cost per life-year gained and incremental cost per QALY gained.

Probabilistic analysis

In a PSA, where the parameters of the survival models (both overall and progression-free survival), health state utility values, treatment cost, cost of outpatient attendances and patient monitoring, as well as costs of managing adverse events and palliative care were sampled probabilistically, radiofrequency ablation is associated with reduced costs (with a range from –£11,189 to –£1369) in all simulations and increased QALYs for the majority of simulations when compared with surgical resection (Figure 16 shows this and also shows the 95% confidence ellipse). However, radiofrequency ablation is associated with lower QALYs in 23% of simulations (with an overall range of –0.21 to 0.07).

FIGURE 16.

Cost-effectiveness plane – incremental cost and incremental QALYs for RFA and surgical resection for (resectable) liver metastases < 3 cm.

The probabilistic analysis generated cost and QALY estimates for each intervention that were similar to those for the base-case analysis (see Table 74 for the base-case analysis). Table 77 reports the mean costs and outcomes from the probabilistic analysis (including the 2.5th and 97.5th percentiles to give an indication of the range of the simulated values) and the ICER for radiofrequency ablation compared with surgical resection, based on the mean values generated in the probabilistic analysis.

TABLE 77 - Costs and outcomes from probabilistic analysis for radiofrequency ablation and surgical resection for (resectable) liver metastases < 3 cm

RFA, radiofrequency ablation.

Procedure	Discounted costs (£)			Discounted QALYs			ICER (£)
	Mean	2.5th percentile	97.5th percentile	Mean	2.5th percentile	97.5th percentile
Surgical resection	25,521	16,749	35,217	3.00	2.32	3.82
RFA	19,192	9,767	28,751	3.07	2.36	3.85	–102,415

In addition to graphing the incremental cost and incremental QALYs for radiofrequency ablation, a cost-effectiveness acceptability curve was derived, representing the proportion of simulations where radiofrequency ablation is cost-effective for a range of willingness-to-pay thresholds, up to £100,000 (Figure 17). In this analysis radiofrequency ablation had a probability of being cost-effective of 100% at a willingness-to-pay threshold of £20,000 per QALY and 100% at a willingness-to-pay threshold of £30,000 per QALY.

FIGURE 17.

Cost-effectiveness acceptability curve for radiofrequency ablation for (resectable) liver metastases < 3 cm.

Probabilistic analysis

In a PSA, where the parameters of the survival models (both overall and progression-free survival), health state utility values, treatment cost, cost of outpatient attendances and patient monitoring, as well as costs of managing adverse events and palliative care were sampled probabilistically, radiofrequency ablation for patients with solitary liver metastases ≥ 3 cm is associated with reduced costs for the majority (97%) of simulations (with an overall range from –£10,721 to £4591) and reduced QALYs for all of the simulations (with a range of –1.63 to –0.63) when compared with surgical resection (Figure 18 shows this and also shows the 95% confidence ellipse).

FIGURE 18.

Cost-effectiveness plane – incremental cost and incremental QALYs for radiofrequency ablation and surgical resection for (resectable) liver metastases ≥ 3 cm.

The probabilistic analysis generated cost and QALY estimates for each intervention that were similar to those for the base-case analysis (see Table 78 for the base-case analysis). Table 81 reports the mean costs and outcomes from the probabilistic analysis (including the 2.5th and 97.5th percentiles to give an indication of the range of the simulated values) and the ICER for radiofrequency ablation compared with surgical resection, based on the mean values generated in the probabilistic analysis.

TABLE 81 - Costs and outcomes from probabilistic analysis for radiofrequency ablation and surgical resection for (resectable) liver metastases ≥ 3 cm

RFA, radiofrequency ablation.

Procedure	Discounted costs (£)			Discounted QALYs			ICER (£)
	Mean	2.5th percentile	97.5th percentile	Mean	2.5th percentile	97.5th percentile
Surgical resection	19,267	13,464	29,456	3.23	2.36	4.11
RFA	15,307	8306	25,301	2.02	1.35	2.82	3259

In addition to graphing the incremental cost and incremental QALYs for radiofrequency ablation, a cost-effectiveness acceptability curve was derived, representing the proportion of simulations where radiofrequency ablation is cost-effective for a range of willingness-to-pay thresholds, up to £100,000 (Figure 19). In this analysis, radiofrequency ablation had a probability of being cost-effective of 0% at all willingness-to-pay thresholds of above £9000 per QALY.

FIGURE 19.

Cost-effectiveness acceptability curve for radiofrequency ablation for (resectable) liver metastases ≥ 3 cm.