Systematic review and economic modelling of the relative clinical benefit and cost-effectiveness of laparoscopic surgery and robotic surgery for removal of the prostate in men with localised prostate cancer

Technical description

Laparoscopic prostatectomy

Experience in gall bladder and kidney surgery highlighted the advantages of a laparoscopic approach to intra-abdominal organ removal. Insufflation of the abdominal cavity and use of endoscopic lens and digital camera systems for image magnification greatly enhanced surgical view, aiding accurate dissection, and reduced bleeding. Technological development in instrument design and the use of differing energy sources for haemostasis added further potential benefits over open surgery. Appreciation of these advantages led to the first series of men undergoing laparoscopic radical prostatectomy being reported in 1997. 22

For standard laparoscopic radical prostatectomy the patient is anaesthetised and positioned supine on the operating table with legs abducted. Following skin cleansing and draping, the abdomen is punctured with a trocar at the umbilicus under vision using a Hassan technique and a pneumoperitoneum induced with CO₂ gas, which is then maintained throughout the operation at a pressure of 10–12 mmHg. A telescopic camera is then inserted though the insufflation port (10 mm diameter) and a further three 5-mm ports and one 12-mm port are inserted in a specific configuration to allow ergonomic access to the pelvis without instrument clashes (Figure 2). The operating table is then adjusted with the patient in a 45° head-down position. The principal operating surgeon then proceeds with dissection of the prostate under televisual control using long narrow instruments such as a diathermy knife, scissors, graspers and needle holders passed through the ports while one or two assistant surgeons maintain the magnified view projected on two television screens by manipulating the telescopic camera and removing blood and fluid by suction. 23 Alternatively, the camera can be operated by a single active robotic manipulator arm that is controlled through voice commands from the operating surgeon. 24 Generally, blood loss is prevented by securing visible blood vessels with clips, diathermy and the use of other energy devices such as ultrasound. By considering preoperative findings and direct inspection of the prostate the surgeon will decide whether to preserve one or both neurovascular bundles attached to the posterolateral surface of the prostate that supply the urinary sphincter and penile erectile tissue. Once the prostate is dissected free it is placed in a retrieval bag within the abdomen and the continuity between the bladder and urethra restored by anastomosis using up to six interrupted sutures or by single continuous suture; a urinary catheter is then placed. One of the 12-mm ports is widened slightly to allow retrieval of the excised prostate, which is sent for pathological examination, haemostasis is then confirmed and the port sites closed with sutures. Anaesthesia is then reversed and the patient transferred to the recovery area for initial observation. The procedure typically takes 3.5–4 hours of operating theatre time. Increasing experience with the technique has demonstrated that it does result in reduced blood loss and earlier return to full activity compared with open prostatectomy, but any reduction in rates of erectile dysfunction and incontinence remains uncertain. 25,26

FIGURE 2.

Configuration of differently sized abdominal port sites through which instruments are introduced for laparoscopic prostatectomy. 23 Reproduced with permission from the International Brazilian Journal of Urology.

Robotic prostatectomy

A surgical robot can be defined as a powered device with artificial sensing that can be programmed or externally controlled by a surgeon to position and manipulate instruments to undertake surgical tasks. The key surgical benefits of robotic technology are to tirelessly make precise repetitive movements to move, locate and hold tools and to respond quickly to changes in commands. Robots are intended to assist rather than replace the surgeon, who retains control at all times. They can be broadly classified into three groups: passive, active and master–slave telemanipulators. 27,28 Early positive experience with passive devices, such as frames to accurately position instruments during brain surgery, and active devices programmed to respond to voice- or pedal-activated commands, such as extra ‘arms’ to position the endoscopic camera during standard laparoscopic surgery, led to the design of master–slave surgical manipulators. Here, the surgeon sits at a master console in the operating theatre separate from the patient and remotely controls arms that position and operate the camera and tools inserted into the patient through ports. The control mechanism can be through a joystick, pedals or, more appropriately for surgery, gloved handles that mimic the movements of the slave manipulator. The technology allows the scaling of motion whereby the relatively gross hand movements of the surgeon are translated to micromotions of the robotic arms. This is further enhanced by ‘wrists’ built into the instruments that allow six degrees of freedom of movement, which more closely approximates the range of movements possible by the human hand during open prostatectomy, rather than the more limited four degrees of freedom possible with standard laparoscopic instruments. An advanced camera lens system allows three-dimensional vision and 10–15 × magnification to be transmitted to the master console. Such master–slave telemanipulators were initially developed from previous US military designs by two commercial companies and used for coronary artery bypass surgery,29 but a subsequent commercial merger resulted in a single company, Intuitive Surgical Incorporated (Sunnyvale, CA, USA), which developed the da Vinci® system for wider clinical use. 30

The advantages of the multi-armed robotic telemanipulator system in terms of improved dexterity of operation of laparoscopic instruments by increasing articulation and scaling together with the three-dimensional magnified image all set in an ergonomic platform encouraged a number of centres, particularly in the USA, to apply this system to radical prostatectomy. It was also thought that the greater scope for telemedicine mentoring and the ability of the robot to scale surgeon movements and hence reduce unwanted movements such as tremor would widen the group of surgeons who could achieve competency at keyhole prostatectomy. 31,32

The initial preparation for robotic prostatectomy is identical to that for the standard laparoscopic procedure. The operating theatre is required to be of a minimum size to accommodate the extra equipment, although this is now standard for newer hospital facilities, including those within the UK NHS. Once the ports (generally six) are placed and the patient tilted in a 45° head-down position, the robot is then ‘docked’ to the patient, which generally takes 15–20 minutes. The docking requires the attachment of one robotic ‘slave’ arm to the telescopic camera while the other two (for the three-arm model) or three (for the four-arm model) are attached to the operating instruments that will be manipulated remotely by the lead surgeon. The arms are housed on a cart that is positioned adjacent to the patient. The assistant surgeon generally operates the suction device or retracting instruments through the remaining ports. The operating surgeon sits at a teleconsole within the operating theatre linked to the robot by cable, although more remote wireless locations are possible (Figure 3). 33 The console comprises a three-dimensional display monitor for the camera-fed operative view, ‘master’ arms linked to the ‘slave’ arms, which allow the surgeon to direct and operate the instruments, camera-positioning controls, foot pedals controlling diathermy for haemostasis and finally a central processing unit to regulate the system. Additional controls can adjust the display, the offset angle of the telescopic camera lens and the ratio of the scaling of surgeon’s movements to instrument movements. The procedure typically takes 3.5–4.5 hours of operating theatre time. Robotic prostatectomy also results in reduced blood loss and quicker return to full activity but again the hoped-for reduction in rates of incontinence and erectile dysfunction as a result of improved vision remains uncertain. 34 A deficiency of the robotic technique is the lack of transmission of the feel of the tissues from the remote instruments; reproduction of this haptic sense is a key aim of future development.

FIGURE 3.

da Vinci surgical robot system showing, from left to right, surgeon at remote console; three-armed (labelled 1–3) telemanipulator for docking to patient; and assistant adjusting room monitor. ©[2011] Intuitive Surgical, Inc. Reproduced with permission from ©2010 Intuitive Surgical, Inc.

Reproduced with permission from ©2010 Intuitive Surgical, Inc.

It should be noted that the robotic technology within the da Vinci system continues to evolve and advancements tend to be added by Intuitive Surgical as options to the basic platform at extra cost. Currently, purchasers of the system can choose to have a fourth robotic arm, reducing the number of surgical assistants required, more advanced image transmission and an additional console to allow mentoring of surgeons under training (similar to dual controls for a motor car).

Requirement for radical treatment of prostate cancer in the UK NHS

In the UK prostate cancer is generally detected by PSA testing of men complaining of lower urinary tract symptoms, although the numbers of asymptomatic men requesting a PSA test to assess their risk of having or developing prostate cancer is increasing, particularly among more affluent socioeconomic groups in the south of England. 35 For men with a serum PSA above a diagnostic threshold currently set in the UK at 3 ng/ml for men in their 50s, 4 ng/ml for those in their 60s and 5 ng/ml for those in their 70s, prostate biopsy is recommended. 8,36 Biopsy involves obtaining 10–12 cores of prostate tissue measuring 10 × 2 mm by transrectal ultrasound (TRUS)-guided needle biopsy as an outpatient procedure under local anaesthetic. This procedure is uncomfortable and is often associated with mild adverse effects such as bleeding and urinary tract infection (30–80%); more severe adverse effects such as systemic sepsis are uncommon (< 1%). 37

At present, approximately 25% of men with PSA levels above threshold will have cancer detected on biopsy,38 with 37,051 men being registered with the diagnosis in the UK during 2008. 11 Following diagnosis a treatment decision has to be made, which will involve consideration of the PSA level, the clinical stage of the cancer categorised on the tumour, node, metastasis (TNM) staging system,39 the aggressiveness of the cancer classified by grading the degree of disruption of the normal glandular architecture of the prostate seen on microscopic examination using the Gleason score40 and person factors such as life expectancy and treatment preference. 12,41,42 For men with apparent localised disease confined to the prostate gland (preoperative clinical classification of tumour stage cT1 and cT2, N0, M0), radical treatment by either surgery or radiation is an option, together with active surveillance programmes, with deferred treatment for men with a Gleason score ≤ 6. 43 Current evidence suggests that any benefit to the individual receiving radical treatment for prostate cancer takes at least 10 years to accrue and therefore these options are best used for men whose comorbidity and age suggests a life expectancy of > 10 years. 44 Finally, evidence is increasing that more aggressive cancers, categorised by a Gleason score of ≥ 8 out of 10 and a PSA of > 20 ng/ml, are likely to already have developed metastases and therefore such patients are considerably less likely to benefit from radical treatment alone. 45 The typical man who undergoes radical prostatectomy therefore is generally fit [American Society of Anesthesiologists (ASA) grade 0–2] and aged < 70 years and has tumour characteristics suggesting low or intermediate risk of disease progression according to the D’Amico risk classification system (Table 1). 46

Estimated demand for radical prostatectomy

Assuming that 45% of men diagnosed with prostate cancer in the UK are aged < 70 years11 and that the disease is localised to the prostate in 86% of cases,47 approximately 14,000 men would have the option of radical treatment each year. Health episode statistics recorded for NHS England48 show that approximately 4000 (28% of the estimated total) men underwent radical prostatectomy in the year 2009–10, this being a similar proportion to that seen for men diagnosed with cancer in the control arm of the European Randomised Study of Screening for Prostate Cancer [946/3402 (28%)]. 49 [It is noted that there is a discrepancy between differing NHS datasets in the numbers of men coded as having a radical prostatectomy in NHS England in the financial year 2009–10: 4100 using the Office of Population Census and Surveys (OPCS) four-character procedure codes compared with 4703 using Healthcare Resource Group (HRG) codes.] The remaining men chose alterative treatment options such as implantation of radioactive seeds (brachytherapy, 15%), external beam radiotherapy (40%) or decided on an active surveillance protocol (17%). Demographic trends in terms of the increasing number of men at risk together with an anticipated continued rise in the use of PSA testing in the UK suggest that the demand for prostatectomy and other options to treat localised prostate cancer will increase over the next 10 years. Using the hypothetical scenario of increased ‘on demand’ use of PSA testing up to the rate currently practised in the USA would give an estimated figure of 7000 men per year,50 and this would rise further to an estimated 11,000 men per year with the hypothetical scenario of a national programme of PSA screening. 49,50

Current use of technologies in UK NHS

Under the NHS Cancer Plan pelvic cancer surgery, including radical prostatectomy, is concentrated within 60 UK cancer centres, of which approximately 20 perform at least some procedures laparoscopically [personal communication from expert panel members (D Neal, C Eden, R Kodelburg, N Soomro, A McNeil), 2010]. In 2010, 16 had access to a da Vinci robotic system, although most robotic systems in the UK were installed in 2009–10 and were not yet fully operational at the time of carrying out this review (Figure 4). 30,51 NHS England reference cost data recorded 1816 laparoscopic/robotic procedures in the year 2009–10, suggesting that these options were used for 46% of all radical prostatectomies. 52 Our own survey of cancer units known to be carrying out laparoscopic and robotic radical prostatectomies suggests a current 50 : 50 split between laparoscopic and robotic techniques, meaning that approximately 23% of radical prostatectomies carried out in the UK at present are performed using the robotic technique. Other areas of the world have experienced a greater uptake of robotic prostatectomy, for example in the USA it was estimated that 43% of all radical prostatectomies were performed using the robotic technique in the year 2006–7 and approximately 70% in 2008. 17,53,54

FIGURE 4.

UK sites with an installed da Vinci robotic surgical system in 2010.

Current costs for the UK NHS

NHS reference costs for England for the financial year 2009–10 published by the UK government’s Department of Health show an average tariff for open radical prostatectomy (HRG code LB21Z) of £4614 with 2897 procedures claimed by NHS hospitals giving a total annual cost of £1,336,758. For laparoscopic and robotic prostatectomy (HRG code LB22Z), the average tariff was £5257, with 1816 procedures claimed, giving a total annual cost of £9,546,712. (It is noted that there is a discrepancy between differing NHS datasets in the numbers of men coded as having a radical prostatectomy in NHS England in the financial year 2009–10: 4100 using OPCS four-character procedure codes compared with 4703 using HRG codes.) These data suggest a grand total tariff-based cost to the English NHS of £10,883,470 for the year 2009–10. Both an increase in the number of radical prostatectomies required and an increase in the proportion of procedures carried out using a laparoscopic or robotic technique would substantially increase the cost to the NHS. For example, a scenario of increased use of PSA testing leading to a demand for 7000 procedures per year that were all carried out laparoscopically or robotically would increase the tariff-based cost by 240% to £36,799,000.

Patient characteristics

The population of patients considered for this review are men with localised prostate cancer undergoing radical prostatectomy at designated pelvic cancer surgical treatment centres within the UK NHS. The patient variables that define this population include age and comorbidity that together determine an estimated life expectancy of at least 10 years. The great majority of such men are able to undergo radical prostatectomy by either standard laparoscopic or robotic techniques; the few exceptions suited only to the open approach are those with poor respiratory reserve, morbid obesity or previous extensive pelvic surgery.

Disease factors are focused on the estimated risk of developing recurrent disease from metastases not identified at preoperative assessment or because of failure to completely remove localised disease. The approximate magnitude of this risk for an individual man diagnosed with prostate cancer can be calculated using a nomogram developed from linear regression models, the most commonly used version being hosted by the Memorial Sloan Kettering Cancer Institute in web-based form. 57 These models use the preoperative disease factors of age, PSA, clinical tumour stage, Gleason grade and number of needle biopsy cores positive for cancer.

Staging of prostate cancer

The stage of an individual’s cancer is categorised according to the Union for International Cancer Control (UICC) 2009 classification (Table 2). 39 Preoperatively this is determined by clinical assessment using digital rectal examination and imaging with the allocated tumour stage (T) given the prefix ‘c’, for example cT1. Following prostatectomy, pathological examination of the prostate and, in some cases, adjacent lymph nodes may result in a change in the staging as more accurate information concerning the size of the tumour and whether it has breached the external surface of the prostate will be available. To indicate this more accurate evaluation, the T stage assigned following pathological examination of the whole prostate is given the prefix ‘p’, for example pT2a. Rarely, no tumour will be found on pathological examination of the prostate following radical prostatectomy for biopsy-proven cancer; this is designated pT0.

Gleason grading

The qualitative low-magnification microscopic histological description of prostate cancer first suggested by Gleason58 remains an essential aspect of prognostic categorisation although there have been substantial modifications over the subsequent years. 40 The classification grades individual areas of prostate cancer according to the degree of disruption of normal glandular architecture, with grade 1 indicating minimal disruption, grade 5 complete loss of normal glandular arrangement and grades 2, 3 and 4 intermediate between these two extremes. Standard practice consists of identifying the first and second most prevalent patterns within a set of biopsy cores, which give the primary and secondary Gleason grades (each rated 1–5). These are then added together to give the overall Gleason sum score (2–10). Recent consensus tends to limit the use of grades 1 and 2 and therefore scores generally range between 6 and 10. 59 Any tertiary higher disease areas are also reported irrespective of their extent. Higher individual grade and total sum score indicate more aggressive disease with the primary grade being more predictive. For example, an individual whose tumour is categorised as Gleason score 4 + 3 = 7 will tend to have a worse prognosis than an individual with a Gleason score of 3 + 4 = 7. 60 Recent consensus mandates that pathological reporting of prostate cancer using the Gleason grading system should include the most prevalent pattern (primary grade), the second most prevalent pattern (secondary grade) and the presence of any areas that are assigned a higher grade than that assigned to either the primary or secondary patterns (tertiary grade). For needle biopsies the Gleason score is obtained by summing the higher of the secondary or tertiary grades. For radical prostatectomy specimens the Gleason score is obtained by summing the primary and secondary grades, any higher-grade tertiary pattern being stated separately if it occupies < 5% of the tumour.

Cancer extent

There is some evidence that the tumour extent on needle core biopsy estimated by measuring the number of cores positive for cancer, the percentage of needle core tissue affected by cancer and the length in millimetres of the core segments with cancer present is also an independent prognostic factor predictive of future disease progression. 61 Similarly, the total volume of cancer identified by pathological examination of the whole prostate after radical prostatectomy has been assessed as a possible predictive factor for recurrence but was found not to be independently significant on multivariate analysis. 62 These pathological measures of cancer extent have not been included in our care pathway given the current uncertainty of the evidence base.

Summary

Variables collected preoperatively for men undergoing radical prostatectomy including age, tumour stage, Gleason score and tumour volume can predict the risk of disease progression at some time after surgery, with stage and Gleason sum score being most useful. It is therefore important that studies comparing treatments, such as this review, include an assessment of whether or not the patient groups undergoing each procedure are balanced for these variables.

Introduction

For the purposes of this review it is assumed that the procedures being considered will be carried out in hospitals that have the necessary resources in terms of staff, facilities and NHS cancer plan approval to carry out either laparoscopic prostatectomy or robotic prostatectomy on a routine basis. This will comprise operating theatre and recovery facilities including critical care and standard urology wards, the required clinical and technical expertise including surgeons, anaesthetists, theatre nursing team, pathologists and technicians, and continued care including outpatient review, repeat imaging and facilities for further treatment for adverse events or cancer progression. The procedures have been described in Chapter 1. 30,63 For the safe conduct of both procedures it is important that all members of the operating theatre team have had specific training in the performance of the procedures, this being particularly crucial from a technical point of view for the robotic procedure.

Surgeon learning curve

Both laparoscopic and robotic prostatectomy are currently being implemented in the UK NHS, requiring the training of surgeons to perform the procedures. The performance of repeated tasks tends to improve with experience and this improvement is characteristically rapid at first and then slower as a steady state expert level is reached, leading to the use of the term ‘learning curve’ to describe the process. Learning of surgical procedures can be additionally influenced by the previous experience of the surgeon or surgical team, case-mix selection, use of multiple outcomes defining ‘success’ and continued development of the technology. 64 The learning curve effect is often crudely quantified by the number of procedures required to reach competence or the reducing time taken to perform the procedure; in open prostatectomy, for example, experience-related changes in performance may continue even after 250 procedures. 18 As use of laparoscopic prostatectomy increased it was realised that the procedure was difficult to master, requiring a high number of training procedures to achieve competence, and that the skills required did not translate directly from those used in open surgery. 65 This is a particular problem in countries such as the UK, where few centres undertake more than 50 cases per year, the suggested volume required for training and maintenance of competency. 66 Findings from individual case series suggest that robotic prostatectomy reduces the number of cases required for competence, enabling the surgeon to reach an expert level quicker, and that previous experience of laparoscopic prostatectomy is not essential. 67 In addition, it is possible that some surgeons who are unable to master the laparoscopic technique can take advantage of the greater movement control offered by the robotic system to become competent in robotic prostatectomy. Any evaluation of effectiveness and safety of the prostatectomy procedures must therefore balance the relative effects of the learning curves.

Pelvic lymphadenectomy

Men whose disease is characterised preoperatively as intermediate or high risk (see Table 1) may be advised to undergo pelvic lymphadenectomy as part of their laparoscopic or robotic radical prostatectomy in order to detect occult lymph node metastases. The lymphadenectomy is performed as the first part of the radical prostatectomy procedure using a standard dissection template and the package of lymph nodes is removed separately from the prostate for subsequent pathological examination. The prostatectomy would be aborted only if there was gross visible lymph node enlargement, which, given preoperative imaging, is a very rare circumstance. For the purposes of this evaluation we chose, in consultation with the expert panel, to assume that all men with intermediate- or high-risk disease undergoing laparoscopic or robotic prostatectomy would also have a pelvic lymphadenectomy. This is in line with current guidance but we do acknowledge the controversy in this area. 45

Hospital stay

Men are generally admitted to hospital either on the day of surgery or the evening before. A rectal enema is administered to clear the lower bowel. Just before surgery prophylactic antibiotics are given according to local policy and venous thrombosis/embolism prophylaxis also commenced. After surgery the patient is routinely nursed on a standard ward in the UK although specific comorbidities or intraoperative complications may require a period in a critical care area. In the UK, men are typically discharged home after 3 days with an indwelling catheter although this can be reduced by managed care programmes. They then return to the ward after a further 7–14 days according to local protocol as a day patient for urinary catheter removal and voiding check.

Perioperative adverse events

Pathological examination of the prostate

Careful and thorough microscopic examination of the removed prostate by an experienced pathologist is required to determine the true extent of the disease and to identify whether or not the surgery may have been unable to remove all of the contained cancer (positive margin), whether or not the cancer has spread outside the prostate (extraprostatic extension) and, if lymphadenectomy has been performed, the presence of lymph node metastatic disease. In addition, a more comprehensive assessment of the Gleason patterns within the cancer is possible. 71 This examination will recategorise the disease according to stage and, if appropriate, lymph node status (pT and pN) and postoperative Gleason sum score, which will allow more accurate estimation of prognosis according to available post-radical prostatectomy prognostic nomograms57 and inform whether early additional (adjuvant) treatment should be advised. The crucial nature of this examination has led to regular international plenary meetings of expert pathologists who have made consensus recommendations guiding best practice for specimen collection, processing, examination and analysis in order to promote consistency in pathologist reporting of radical prostatectomy specimens. 59,72

Follow-up schedule

Men who have undergone radical prostatectomy are generally seen by the operating team as outpatients 6 weeks after their surgery and then 3-monthly for the first year and 6-monthly for the next 4 years. At each follow-up consultation serum PSA is checked for evidence of tumour recurrence and a qualitative assessment made for continence and desired sexual function. If further assessment or treatment is required for any of these aspects then the pathway of care will be changed accordingly (see Figure 5).

Detection of persistent or recurrent disease

The risk of disease recurrence is higher if one or more of the following disease factors are present: preoperative PSA > 20 ng/ml, pathological Gleason score > 7, pathological extraprostatic disease (pT3/pT4), pathological positive margin or positive lymph nodes (pN1/pN2). If positive lymph nodes are found or the likelihood of disease persistence or recurrence is otherwise deemed to be very high then immediate adjunctive treatment may be offered. For the majority of men, however, PSA surveillance is started according to a standard schedule, for example that defined in the preceding paragraph. Following removal of the prostate, serum PSA (half-life 2.2 days) levels will rapidly fall to an undetectable level, defined as values less than the sensitivity of the assay. Generally, ultrasensitive PSA assays are used for men following radical prostatectomy giving postoperative values of < 0.01 ng/ml. Definitions of the threshold of PSA rise that signifies cancer recurrence vary but generally the finding of two successive PSA readings > 0.2 ng/ml is used, this being denoted biochemical recurrence. 73,74 Once biochemical recurrence occurs a decision will be made with the patient whether to continue surveillance or commence adjuvant treatment. This decision will be informed by tests such as magnetic resonance imaging and radionuclide bone scanning designed to demonstrate the site of recurrence as being in the prostatic bed (localised) or as lymph node or bony metastases (systemic).

Adjuvant treatment

For purely localised recurrence radical radiotherapy is recommended as defined in the RADICALS trial protocol. 75 The treatment consists of delivery of up to 66 Gy of radiation divided into daily doses over 4–6 weeks. It is uncertain whether or not the addition of short-term androgen deprivation is beneficial for presumed localised disease, a research question that RADICALS is designed to address. For men with likely systemic recurrence, long-term, typically life-long, androgen deprivation therapy (medical castration) most commonly achieved with a luteinising hormone-releasing hormone (LHRH) agonist is recommended. This consists of 3-monthly subdermal injections of a depot preparation of the chosen drug. Alternatively, some men may choose surgical castration, removing both testicles (bilateral orchiectomy). The use of long-term androgen deprivation therapy or bilateral orchiectomy for metastatic disease is thought to be palliative because at some point the disease will lose androgen dependency (castrate-resistant prostate cancer). The duration from start of therapy to escape from androgen control, signified by a further substantial rise in PSA values, varies according to the aggressiveness and extent of disease, with a median time of approximately 12 months. Side effects of androgen deprivation therapy include hormonal changes leading to hot flushes, gynaecomastia and altered fat distribution together with osteoporosis. Men with castrate-resistant prostate cancer have a median survival of approximately 18 months and further treatment is usually palliative with symptom control and use of corticosteroid drugs to improve well-being. The chemotherapeutic agent docetaxel does have some activity, extending survival by 3 months on average, but is suited only to men with good performance status. 76

Urinary incontinence

Recovery of continence following radical prostatectomy can take up to 12 months although most men will regain continence by 6 months. In general, therefore, men suffering urinary incontinence will be advised to use containment devices such as absorbent pads or penile sheath drainage for the initial 12 months. If bothersome leakage persists beyond this time then the main treatment options will be surgical implantation of an artificial urinary sphincter (AUS) or continued use of containment devices. For the purposes of this evaluation we used the individual definition of urinary incontinence given in each study without attempting to separate out differing definitions or categorisation of severity. A recently reported randomised controlled trial (RCT) of pelvic floor muscle therapy following radical prostatectomy demonstrated that the rate of urinary incontinence beyond 12 months using patient-reported measures and data collection independent of the clinical team was higher than that given by most of the studies used in our meta-analysis. 77

Erectile dysfunction

For men who were sexually active before surgery, approximately 40% will experience worsening of their sexual function and in particular difficulty initiating and sustaining penile erection sufficient for desired sexual activity. This is particularly dependent on preservation of one or both neurovascular bundles at the time of radical prostatectomy. Similar to urinary incontinence full recovery can take up to 12–18 months following surgery. For men with persistent and bothersome erectile dysfunction, treatment options will include drug treatment taken on an as-required basis, a vacuum constriction device or penile implant surgery. Most men will first trial an oral phosphodiesterase type V inhibitor, with a suggested prescribing frequency of one treatment per week according to NHS guidance. The next option will be alprostadil (Carerject®, Pfizer) given as an intraurethral pellet or an intracavernosal injection with suggested NHS prescribing frequency again of one treatment per week. For men who achieve satisfactory restoration of sexual activity with these drugs their use will continue long term. If drug treatments are unsuccessful men may trial a vacuum constriction device or consider surgical implantation of a penile prosthesis. The proportion of men pursuing these last two options is small as most will accept their loss of sexual function in the longer term. In addition, it should be noted that, although this outcome is an important aspect determining treatment selection for many men with localised prostate cancer, the definition of any deterioration is not standardised and collection of data concerning sexual function before and after surgery is generally poor. Most studies do not separately categorise those men who were sexually active before surgery and who underwent deliberate nerve-sparing surgery with the aim of preserving sexual function.

Number of studies identified

The searches identified 2722 potentially relevant titles and abstracts (Figure 6), from which 914 reports were selected for full-text eligibility screening. Of these, 58 reports (54 studies) were included and 856 reports were excluded with reasons for exclusion detailed in Figure 6. We attempted to obtain further details for 69 of the 80 (86%) reports that were excluded because of lack of clear information on the number of patients for each baseline clinical stage and which had contact details available. Nineteen replies were obtained. Only one of these 19 reports89 was subsequently deemed eligible for inclusion, but confirmation of this was received too late for it to be included in the review. Appendices 5 and 6 give the bibliographic details of the included and excluded studies respectively.

FIGURE 6.

Flow chart of the number of potentially relevant reports of identified studies and the numbers subsequently included and excluded from the clinical effectiveness review.

Number and type of included studies

The searches identified one RCT of laparoscopic versus open radical prostatectomy90 and 57 non-randomised comparative reports of 53 studies from 40 different clinical institutions: eight robotic versus laparoscopic prostatectomy;91–98 four robotic versus laparoscopic versus open prostatectomy [three primary,99–101 one secondary102 (earlier report of the same study but containing unique data)]; 18 robotic versus open prostatectomy (16 primary,103–118 two secondary119,120) and 27 laparoscopic versus open prostatectomy (26 primary,121–146 and one secondary147). There were three conference abstracts: two comparing robotic versus laparoscopic prostatectomy94,97 and one comparing robotic versus laparoscopic versus open prostatectomy. 102 Four studies were considered to include potential patient overlap: the study conducted by Menon and colleagues95 was a comparison of 40 laparoscopic and 40 robotic prostatectomies performed between 23 October 2000 and 22 October 2001; Tewari and colleagues116 report an extension of this work but compared 100 open and 200 robot operations between October 1999 and December 2002. As these studies included different comparators, they were treated as separate studies but the potential for overlap of robotic prostatectomy patients was noted. Similarly, Joseph and colleagues94 report a comparison including 800 laparoscopic cases from the Henri Mondor hospital, France, and 745 robotic cases from the University of Rochester, USA, between 2002 and 2006. An earlier publication93 analysed the last 50 cases from a series of 70 laparoscopic and 200 robotic cases from the University of Rochester (dates not given). The studies were treated as separate. Similar affiliated institution details of first authors were noted for seven studies: those by Anastasiadis122 and Salomon,140 Ficarra106 and Fracalanza,107 and Greco,129 Jurczok131 and Fornara. 127 These studies report overlapping treatment dates and similar procedures but it is unclear whether or not they include patient overlap as details of the institutions where the men were treated are not clearly given within the reported text. Similarly, we noted similar author institution details for another seven studies: those by Malcolm,110 Ball99 and Soderdahl,142 Trabulsi98 and Brown,125 and Loeb109 and Wagner146 although these involved different comparison groups and were treated as separate studies.

The 57 non-randomised comparative reports (of 53 studies) included 28 prospective and 17 retrospective reports. Three studies92,112,114 included a mixture of prospective and retrospective data and eight96,97,100,119,123,132,134,138 did not report the method of data collection. The method of data collection was uncertain in the study by Kim and colleagues132 because of a limited translation of the full-text version. Table 4 provides further details of the number and type of included studies.

The RCT conducted by Guazzoni and colleagues90 comparing laparoscopic with open prostatectomy was set in Italy. Half of the included non-randomised studies were conducted in the USA (28/57, 49%). The remaining studies were conducted in France,91,94–96,101,122,140 Italy,106,107,114,123,129,134 Germany,127,131,137 Japan,135,136,144 Canada;121,130 there was one study from each of Australia,105 Austria,139 Brazil,141 Chile,133 Croatia,143 Republic of Korea,132 Spain,138 Sweden104 and Taiwan, Province of China. 113 Of the non-randomised comparative studies comparing robotic with laparoscopic radical prostatectomy, three primary full-text studies92,93,98 and one conference abstract97 were set in the USA, one conference abstract was set in both the USA and France94 and three studies were set in France. 91,95,96 Of the non-randomised comparative studies comparing robotic, laparoscopic and open radical prostatectomy, two primary studies99,100 and one secondary report102 were set in the USA and one study was set in France. 101 Of the non-randomised comparative studies comparing robotic and open radical prostatectomy, 10 primary studies103,108–112,115–118 and two secondary reports119,120 were set in the USA, one study was set in Australia,105 three primary studies106,107,114 were set in Italy, one study was set in Sweden104 and one was set in Taiwan, Province of China. 113 Of the non-randomised comparative studies comparing laparoscopic and open radical prostatectomy, seven primary studies124–126,128,142,145,146 and one secondary report147 were set in the USA, three primary studies127,131,137 were set in Germany, three primary studies135,136,144 were set in Japan, three primary studies123,129,134 were set in Italy, two primary studies122,140 were set in France and one study each was set in Austria,139 Brazil,141 Canada,121 Chile,133 Croatia,143 Republic of Korea132 and Spain. 138

The four full-text publications that required translation paired with their original language were Fornara and Zacharias127 (German), Kim132 (Korean), Soric143 (Croatian) and Raventos Busquets and colleagues138 (Spanish).

Characteristics of patients

The 58 reports included 21,126 men at enrolment. Excluding secondary reports and following exclusions because of ineligibility or participant dropout, the final study analyses included 19,064 men, of whom 6768 underwent robotic radical prostatectomy, 4952 underwent laparoscopic radical prostatectomy and 7344 underwent open radical prostatectomy. The demographic and disease characteristics of these included men are summarised in Table 5.

All studies reported age with a median (interquartile range) of 62 (60–64) years and a total range of 35–84 years.

Baseline clinical tumour staging data were reported for all studies except that conducted by Bolenz and colleagues;100 however, clinical staging data for this study were available from an earlier report in abstract form. 102 Eight reports107,111,120,126,139,141,143,147 did not report specific baseline clinical stage, simply reporting their inclusion criterion as ‘≤ cT1–T2’, and one109 did not report clinical stage by procedure. The baseline clinical tumour staging was similar between the laparoscopic and robotic radical prostatectomy patients with 68% and 69%, respectively, categorised as T1.

Less than half of the included reports (23/58, 40%)91,98,99,101,103,105–108,110,115,117–121,125,128,135,136,142,145,146 gave detailed biopsy Gleason scores for men undergoing prostatectomy in the format we required: numbers of men categorised as Gleason score ≤ 6, 7 or ≥ 8. Seven studies90,95,97,111,126,139,141 and one secondary report147 did not report biopsy Gleason grades or score. Over one-third of the included reports (21/58, 36%) reported either mean93–95,113,122–124,129,130,132,139,140,143,144 or median104,114,127,131,133,134,137 scores. The remaining reports presented details using different scoring formats90,92,102,138,141 or did not present separately by procedure. 100 Two-thirds of men undergoing both laparoscopic and robotic radical prostatectomy had a Gleason score ≤ 6.

Fifty reports90,91,93–101,103–109,112–119,122–125,127–146 gave preoperative PSA values, with the majority (38/50, 76%) reporting mean PSA for each group of men. Nine studies106–108,131,134,141,142,144,145 reported median group PSA values, whereas two studies135,136 reported mean and median PSA and one study119 reported PSA range only. Combining the median and mean PSA values across all of the studies demonstrated slightly lower levels of preoperative PSA in the robotic than in the laparoscopic procedures: 6.3 ng/ml and 7.2 ng/ml respectively. Three studies92,121,126 reported the number of men in each group falling into varying ranges of PSA values but as the ranges were inconsistent we were unable to include these data in the summary.

The postoperative Gleason sum score following pathological examination of the prostate was similar between the robotic and laparoscopic patients with 50% of the men in both groups with combinable Gleason information having a Gleason score ≤ 6. Pathological staging assigned following consideration of the operative finding during surgery and pathological examination of the removed prostate was similar between the robotic and laparoscopic patients with 78% of the men with combinable staging information in both groups categorised as pT2. There was a trend towards worse disease characteristics in men undergoing open prostatectomy with 55% having a post-prostatectomy Gleason score > 6 and 30% categorised as pT2 or higher.

Twenty-nine primary reports90–93,96,99,100,106,108,110–113,118,122,123,125,126,128,129,132,135–137,139,142,144–146 and two secondary reports102,119 reported the use of nerve-sparing techniques.

Overview of types of outcomes reported

The numbers and types of included studies reporting our main considered outcomes are summarised below.

Risk of bias

Risk of bias for reported outcomes

The risk of bias assessments for our chosen main outcomes of efficacy (predominantly surgical margins status), urinary incontinence and erectile dysfunction and perioperative adverse events are summarised in Figures 7–10 respectively.

FIGURE 7.

Summary of risk of bias assessment for reports of efficacy (n = 37).

FIGURE 8.

Summary of risk of bias assessment for reports of urinary dysfunction (n = 23).

FIGURE 9.

Summary of risk of bias assessment for reports of erectile dysfunction (n = 20).

FIGURE 10.

Summary of risk of bias assessment for reports of perioperative safety (n = 35).

Positive margins

Meta-analysis of data from the 37 included studies90,94–98,101,103,105–109,112–116,118,122,123,125,127,129–134,137, 139–141,143,144,146,147 that reported positive surgical margin rates (Table 6) showed a statistically significant improvement for robotic compared with laparoscopic prostatectomy (OR 0.69; 95% CrI 0.51 to 0.96; probability outcome favours robotic prostatectomy = 0.987). The probability of a positive margin predicted by the mixed-treatment comparison model was 17.0% following robotic prostatectomy compared with 23.6% following laparoscopic prostatectomy. Restriction of the meta-analysis to studies at low risk of bias did not change the direction of effect but did decrease the precision of the effect size (OR 0.73; 95% CrI 0.29 to 1.75), with the probability that the event rate was lower for robotic prostatectomy being no longer statistically significant (p = 0.782).

Biochemical recurrence

Biochemical recurrence rates up to 1 year following radical prostatectomy were reported in six studies (Table 8). 108,113,115,123,133,137 There was no evidence of a difference in the rates of biochemical recurrence calculated by the mixed-treatment comparison model between robotic and laparoscopic prostatectomy (OR 0.89; 95% CrI 0.24 to 3.34; probability outcome favours robotic prostatectomy = 0.588). Restriction of the meta-analysis to only the studies at low risk of bias was not possible because all studies were at high risk.

Urinary incontinence

The 22 studies that reported urinary incontinence used a variety of measures at different time points. Measures included observed urinary leakage,93 pad use,91,97,108,112–114,116,122,128,129,137,139,146 fluid volume voiding diary130 and validated questionnaire scores [University of California Los Angeles – Prostate Cancer Index (UCLA-PCI)99,110,135,136,142 and International Consultation of Incontinence Questionnaire (ICIQ-UI)106]. Artibani and colleagues123 measured both urinary leakage and pad use. The study conducted by Lama and colleagues133 did not give a definition of incontinence. The results from the 10 studies106,108,113,114,126,128–130,133,146 that reported urinary incontinence at a standard time point of 12 months following prostatectomy are given in Table 9. There was no evidence of a difference in the rates of urinary incontinence between robotic and laparoscopic prostatectomy (OR 0.55; 95% CrI 0.09 to 2.84; probability outcome favours robotic prostatectomy = 0.783). Restriction of the meta-analysis to only the studies at low risk of bias was not possible because all studies were at high risk.

The study conducted by Carlsson and colleages104 reported 7/1253 (0.6%) patients requiring further postoperative surgery for incontinence between 30 days and 15 months after their initial robotic operation compared with 11/485 (2.2%) requiring further postoperative surgery for incontinence after undergoing an open radical prostatectomy.

Erectile dysfunction

As described in Overview of type of outcomes reported, a total of 19 studies provided data on sexual function. The time point following surgery when the outcome was assessed and the measure used to quantify the outcome varied between studies. Erectile dysfunction was variously defined as the inability to achieve and maintain a spontaneous or drug-assisted erection suitable for sexual intercourse93,108,113,114,116,122,123,126,129 or by validated symptom questionnaire scores [UCLA-PCI,99,110,135,136,142 IIEF-5106,128 Expanded Prostate Cancer Index Composite, sexual function subscale (EPIC-SFSS)146 and Sexual Health Inventory for Men (SHIM)112]. The study conducted by Lama and colleagues133 did not report a definition of erectile dysfunction. Given the diversity of definitions and types of data (continuous and dichotomous) it was not possible to collate data from individual studies into a form suited to meta-analysis. Of the two studies directly comparing robotic and laparoscopic prostatectomy that reported erectile dysfunction, one99 showed earlier recovery of sexual function following the robotic prostatectomy procedure, with 35% compared with 21% returning to baseline functioning at 3 months post surgery and 43% compared with 25% returning to baseline functioning at 6 months, and the other93 favoured laparoscopic prostatectomy (46% required drug aid vs 36% at 3 months in the robotic and laparoscopic groups, respectively).

Quality of life

Quality of life following prostatectomy as measured by validated patient-reported questionnaires was reported in 10 studies: European Quality of Life-5 Dimensions (EQ-5D) visual analogue scale (VAS); 90,116,139 Short Form questionnaire-36 items (SF-36);135,136 Short Form questionnaire-12 items (SF-12);111 European Organisation for Research and Treatment of Cancer Quality of Life Questionnaire C30 (EORTC QLQ-C30);137 the quality-of-life item contained within the International Prostate Symptom Score (I-PSS);130 the International Continence Society (ICS)91 and the Expanded Prostate Cancer Index Composite urinary incontinence and sexual function subscales (EPIC-UISS-SFSS). 146 Full details are given in Appendix 9. Quality-of-life measurements following robotic prostatectomy were reported by two studies111,116 with a maximum observation period of 6 weeks. The data were insufficient to enable us to assess any difference in quality of life following robotic or laparoscopic prostatectomy. Three studies135–137 reported that preoperative physical functioning level was not achieved in all patients by 6 months postoperatively but the clinical significance of the differences was unclear.

Need for further cancer treatment

Dahl and colleagues126 was the only report that included information on the numbers of men requiring further treatment for cancer persistence or recurrence, with rates of 5/104 (5%) for laparoscopic prostatectomy and 2/102 (2%) for open prostatectomy.

Death

Four studies104,105,126,140 reported deaths resulting from complications in the 30-day postoperative period. These included two fatal cardiac arrests104,126 and one cerebrovascular accident105 following open prostatectomy. Salomon and colleagues140 also reported one death due to pulmonary embolism following laparoscopic prostatectomy. Five studies92,95,96,137,154 involving 1600 men specifically reported no postoperative deaths. Drouin and colleagues101 reported one death due to prostate cancer 5 years after open prostatectomy and four deaths due to cardiovascular complications without specifying which procedure these men had received. Krambeck and colleagues108 reported all-cause mortality rates of 4/248 (1.6%) for men undergoing robotic prostatectomy and 4/492 (0.8%) after open prostatectomy at a median follow-up time of 1.3 years.

Perioperative adverse events

Data on the perioperative adverse events of blood transfusion, anastomotic leak, bladder neck contracture, wound infection, organ injury, ileus, deep-vein thrombosis and pulmonary embolism are presented in Tables 10–17. Abstracted data concerning other specific adverse events not included in the meta-analysis are detailed in Appendix 10. All adverse events were additionally categorised according to the Clavien–Dindo system and the data meta-analysed according to Clavien–Dindo score (see Tables 59–70).

Clavien–Dindo scores

The predicted event rates based on the meta-analysis statistical models for each Clavien–Dindo category are shown in Table 18. The individual study data contributing to each meta-analysis are given in Appendix 9. The OR for each Clavien–Dindo score was in favour of the robotic procedure but only that for Clavien IIIb, adverse event requiring intervention under general anaesthesia, was statistically significant (Figure 11).

TABLE 18 - Predicted rates of event for each Clavien–Dindo score

Clavien–Dindo category (see Table 3)	Robotic (%)	Laparoscopic (%)	Open (%)
Clavien I	2.1	4.1	4.2
Clavien II	3.9	7.2	17.5
Clavien IIIa	0.5	2.3	1.8
Clavien IIIb	0.9	3.6	2.5
Clavien IVa	0.6	0.8	2.1
Clavien V	< 0.1	0.2	0.2

FIGURE 11.

Odds ratio and 95% CrI by Clavien–Dindo score.

Descriptors of care

Model specification: purpose and design

The purpose of this model was to simulate the outcomes and costs during and following a radical prostatectomy procedure using either a robotic or laparoscopic technique performed in an appropriate UK NHS hospital on a man with clinically apparent localised prostate cancer. 43 The model was specified to follow the predefined care pathway for individual men for 10 years from the time of surgery, this being the anticipated duration of use of the current robotic technology under study (Intuitive Surgical, June 2010, personal communication). We also included as a sensitivity analysis the ability to specify the model over the lifetime of the individual, consistent with the epidemiological characteristics of localised prostate cancer, which typically has a long natural history with survival benefits for radical treatment needing at least 10–15 years to accrue. 44

We selected an individual-based event model in which surgical procedure, steps in the care pathway, the occurrence of longer-term adverse events and ultimately death are modelled as discrete events for individuals within the model. 174 The transition of individual men between events was driven by the previous health states, processes involved in their clinical treatment and subsequent care that arose as a consequence of the surgery, the underlying disease and natural lifespan. These included adverse events associated with the prostatectomy, events during clinical management of individuals who were cured of prostate cancer by the surgery and events driven by disease persistence or recurrence following prostatectomy. The clinical characteristics of individuals entering the simulation could be varied to represent the complete spectrum of patient and disease characteristics among the overall population of men with localised prostate cancer requiring radical prostatectomy. Each event and each subsequent patient management decision at all decision points in the pathway was modelled probabilistically based on available data relevant to patient care in the UK NHS. The hierarchy of data sources used was in the order of the associated systematic review, available relevant literature including web-based sources and consultation with relevant experts. The model was parameterised using data obtained from these sources describing disease progression, survival and the prevalence of adverse events. Data on costs to the UK NHS of laparoscopic and robotic prostatectomy were predominantly obtained directly from the manufacturer of the robotic system, Intuitive Surgical,30 and from national and local NHS sources (see Costs). To enable analysis of cost-effectiveness, utility values for the various health states within the care pathway were obtained from the literature (see Utilities). The model was constructed using the scripting language available for the R statistical package for computing. 175

State variables and timescales

Assumptions within the model

Modelled events at each decision point within the pathway were discrete and independent. For example, surveillance for biochemical recurrence was simulated in the same way irrespective of events previously experienced by the individual. In the absence of suitable data the probability of further biochemical recurrence was independent of previous biochemical recurrences that had been successfully treated. In practice, care options inevitably are affected by previous disease characteristics and other related events, but the multitude of possibilities of care for particular individuals during the course of their cancer care subsequent to radical prostatectomy could not be fully parameterised in the model in the absence of sufficiently detailed individual-level data sets. Proportions of individuals undergoing different procedures within the care pathway were defined by the probability of being assigned to those procedures. This simplification was necessary because of the lack of data on the underlying causal factors leading to events; they were therefore modelled as random processes (see Modelling of discrete events).

The imprecision/uncertainty surrounding parameter estimates used within the model was characterised by assigning statistical distributions to parameters. For parameter estimates provided by the systematic review, the log-normal distribution was used to define the degree of surrounding uncertainty. Other parameters derived from the literature or other sources were considered for accuracy, credibility and plausibility at meetings of the expert panel. Identifying a suitable distribution for estimates and describing the uncertainty around these values was problematic. In such circumstances, uncertainty was calculated as a potential range of plausible values of ±25% of the estimate.

For parameters not defined by the systematic review we assumed that the point estimate was the most likely ‘real’ value and therefore did not consider that a uniform rectangular distribution was appropriate. Furthermore, by defining the extreme limits of the distribution using the triangular method (as described above) we ensured that the upper and lower bounds of variability did not exceed clinical plausibility. And finally, the way in which variability was calculated ensured that the degree of uncertainty applied to each intervention equally.

Modelling the care pathway

Following robotic or laparoscopic prostatectomy each individual was entered into the specific pathway dictated by his clinical and disease state after the operation (Figure 15). This state was characterised in terms of, first, cancer status and, second, the presence of one or more of the three adverse events that were deemed to persist beyond the treatment period: bladder neck contracture, urinary incontinence and erectile dysfunction. The individuals then proceeded through a series of events dependent on where they were in the care pathway and which would result in changes to, or resolutions of, differing health states. This would particularly include remission or relapse following additional treatment for recurrent prostate cancer or resolution of a longer-term adverse event by treatment.

FIGURE 15.

Schematic showing care pathway for the perioperative state during and immediately after radical prostatectomy. A0, perioperative health state; A1, postoperative health state; B2, surgeon learning effect; B3, perioperative complication classified using the Clavien–Dindo system; LRP, laparoscopic radical prostatectomy; ORP, open radical prostatectomy; RRP, robotic radical prostatectomy.

Events were modelled probabilistically using data derived from the hierarchy of sources defined previously in Model specification: purpose and design. Where possible the data used were relevant to both the clinical context of radical prostatectomy and current practice in the UK NHS. 43 Parameters, their values, their distributions and their sources are listed in abbreviated form in the relevant sections. Events experienced by individuals were scheduled in interacting ‘streams’. Surveillance, cancer treatment and mortality were first simulated either until the end of the time horizon if the individual survived or until a process within the care pathway led to death either from prostate cancer or from any other cause (see Figure 15). This provided the framework for each individual’s trajectory through the cancer care pathway. The second set of events simulated the management of the three postoperative dysfunctions: bladder neck contracture, urinary incontinence and erectile dysfunction. If a process led to an intervention event, such as surgery for urinary incontinence, this was scheduled only after at least 12 months of surveillance without a cancer-related event.

Modelling of discrete events

All events were assumed to be binomial in the sense that an event either occurred, 1, or did not occur, 0. Simulation of the occurrence of an event for an individual was undertaken by drawing random uniform deviates and comparing the observed deviate with the known probability of that event occurring given the relevant conditions. Thus, if x represents the proportion of men who experienced bothersome erectile dysfunction after laparoscopic prostatectomy, any random deviate drawn for an individual that was less than x would lead to that individual suffering the dysfunction and progressing down the appropriate pathway of care, whereas any deviate greater than x corresponded to no dysfunction. The proportion of men experiencing each event in each pathway was derived where possible from the systematic review reported in detail in Chapter 4. Other relevant literature or expert opinion were used where necessary.

Perioperative state

In line with the objective of this HTA all patients were assumed to have undergone radical prostatectomy by either laparoscopic or robotic means (see Figure 15). In addition, those individuals deemed to be at intermediate or high risk of early biochemical recurrence according to preoperative disease characteristics (Table 22) were allocated to undergo a concurrent pelvic lymph node dissection; the probability of this was defined from an appropriate additional literature source177 as the information was not available from the systematic review. Adverse events during surgery could initiate two further model events. First, the probability of suffering perioperative adverse events, categorised using systematic review data according to the Clavien–Dindo system into one of six levels, was defined as the proportion of patients who suffered that event68,69 (Figure 16). Second, and independently of adverse events categorised by the Clavien–Dindo system, a proportion of men undergoing laparoscopic or robotic prostatectomy were deemed to require conversion to an open procedure because of intraoperative difficulties. The rate for each of the procedures was determined from the systematic review and the consequence in terms of costs was defined as an extra 3-day hospital stay, decided by expert opinion (see Table 22).

FIGURE 16.

Flow chart illustrating the classification of various intraoperative adverse events using the Clavien–Dindo system for the grading of operative complications and their input into the perioperative care pathway. A0, perioperative health state; CRF, Clavien–Dindo risk factor; B3, perioperative complication categorised by the Clavien–Dindo system.

For each specific Clavien–Dindo level or adverse event the associated financial cost was modelled solely through the extra duration of hospitalisation measured in days that a patient would require according to expert opinion (Table 23). These events were assumed to have resolved during the 3-month perioperative state.

Postoperative state

Immediate further cancer treatment

A proportion of men were assigned to require and undergo immediate further cancer treatment; the probability of this occurring was defined according to the findings of the systematic review, other literature sources and consensus of expert opinion (Table 24). First, men who had undergone pelvic lymphadenectomy as part of their radical prostatectomy and were found to have lymph node metastases on pathological examination of the removed lymph nodes were automatically selected for immediate further treatment. 178 The proportion of men who underwent lymphadenectomy and the proportion of those who were positive were assigned independently from other variables according to the observed rates following either type of surgery from literature sources validated by our expert panel. 177,179 Expert opinion deemed that all men with positive lymph nodes were assigned to further cancer treatment without the opportunity for a period of surveillance.

TABLE 24 - Parameter values (base case) with distributions and sources for lymph node metastases (for men undergoing pelvic lymphadenectomy) and positive margin status (all men)

NA, not applicable.

Upper and lower limits of triangular distribution calculated at ±25% of the point estimate. Upper and lower limits of log-normal distribution set at 95% CI.

Perioperative state	Probability	Lower limita	Upper limita	Assigned distribution	Source
Robotic surgery
Positive margin	0.163	0.119	0.225	Log-normal	Systematic review
Lymph node metastases	0.026	0.0195	0.0325	Triangular	Kawakami 2006179
Laparoscopic surgery
Positive margin	0.236	0.080	0.394	Log-normal	Systematic review
Lymph node metastases	0.026	0.0195	0.0325	NA	Kawakami 2006179

Second, men who had two or more of the following features found on pathological examination of the removed prostate were considered for immediate further treatment:

• positive surgical margin

• Gleason score 8–10

• tumour stage pT3–pT4.

If only one of these pathological disease characteristics was present the individual entered the surveillance pathway (Figure 17).

FIGURE 17.

Schematic representation of postoperative care pathway for men being considered for immediate further cancer treatment. A1, postoperative health state; A2, further cancer treatment state; A3, long-term adverse event dysfunction state; A4, surveillance state; BCR, biochemical recurrence.

Parameterisation of this decision-based approach required linked data for individuals concerning the three features and this was not available from the systematic review. We therefore decided on the following approach. Linked values of postoperative Gleason sum score and postoperative tumour stage for 4669 individuals were kindly provided from a large single institutional database of men undergoing radical prostatectomy maintained at the Vanderbilt-Ingram Cancer Center, TN, USA (D Barocas, February 2011, personal communication). The numbers of men from this data set with each combination of Gleason sum score and tumour stage were then multiplied by the probability of men having a negative or positive surgical margin following robotic or laparoscopic prostatectomy defined by the systematic review and meta-analysis (see Table 24). The calculated patient numbers were then converted to percentages of the sample population, which defined the probability of each combination of the three variables (margin, Gleason sum score and tumour stage) for each procedure. These probabilities were then mapped to the decision matrix. The decision matrix, which directed the subsequent care pathway for individual men in the model, was formulated by rounds of consensus building with relevant members of the expert panel. The decision to be made was whether men would enter the surveillance state or proceed to further cancer treatment (Tables 25 and 26). The decision matrix gave total probabilities of 0.098 following robotic prostatectomy and 0.113 following laparoscopic prostatectomy for individual men requiring consideration for immediate further treatment.

TABLE 25 - Immediate further cancer treatment matrix for individuals following robotic prostatectomy according to findings on pathological examination of the removed prostate

Margin status: negative/positive; tumour stage: negative pT1–pT2, positive pT3–pT4; Gleason score: negative ≤ 7, positive 8–10.

Margin status	Tumour stage	Gleason score	Number (%) of men in category	Probability of event in model	Management decision
Negative	Negative	Negative	2900 (62.1)	0.621	Surveillance
Negative	Negative	Positive	132 (2.8)	0.028	Surveillance
Negative	Positive	Negative	612 (13.1)	0.131	Surveillance
Negative	Positive	Positive	264 (5.6)	0.056	Treatment
Positive	Negative	Negative	565 (12.1)	0.121	Surveillance
Positive	Negative	Positive	26 (0.6)	0.005	Treatment
Positive	Positive	Negative	119 (2.6)	0.026	Treatment
Positive	Positive	Positive	51 (1.1)	0.011	Treatment

TABLE 26 - Immediate further cancer treatment matrix for individuals following laparoscopic prostatectomy according to findings on pathological examination of removed prostate

Margin status: negative/positive; tumour stage: negative pT1–pT2, positive pT3–pT4; Gleason score: negative ≤ 7, positive 8–10.

Margin status	Tumour stage	Gleason score	Number (%) of men in category	Probability of event in model	Management decision
Negative	Negative	Negative	2647 (56.7)	0.567	Surveillance
Negative	Negative	Positive	121 (2.6)	0.026	Surveillance
Negative	Positive	Negative	558 (12.0)	0.120	Surveillance
Negative	Positive	Positive	241 (5.2)	0.052	Treatment
Positive	Negative	Negative	818 (17.5)	0.175	Surveillance
Positive	Negative	Positive	37 (0.8)	0.008	Treatment
Positive	Positive	Negative	173 (3.7)	0.037	Treatment
Positive	Positive	Positive	74 (1.6)	0.016	Treatment

Cancer treatment allocation

Men with pathologically involved lymph nodes or with two or more adverse pathological characteristics listed earlier were immediately assigned to the cancer treatment process following prostatectomy (Figure 18). The extent of the likely residual disease was defined as localised or systemic (metastatic) and this was randomly determined according to known probabilities using the same method described in Modelling of discrete events; this was independent of the precise cancer state variables (see Table 27). A similar process was used for men who underwent an initial period of surveillance and then suffered biochemical recurrence.

FIGURE 18.

Schematic diagram for the further cancer treatment care pathway. A2, cancer treatment health state; A3, long-term adverse event dysfunction health state; A4, surveillance health state.

Persistent adverse event states

Introduction

The incidence of the considered postoperative adverse events or dysfunctions – bladder neck contracture, urinary incontinence and erectile dysfunction – was defined according to the standard parameterisation hierarchy described above. Management of these postoperative dysfunctions was modelled by treating them as independent processes. If dysfunctions were found to be present, self-management and/or treatment began immediately according to current clinical practice (Figures 19 and 20). Each of the three dysfunction-related state variables was recorded as a categorical variable encoding the presence or absence of the pathological condition. These three variables were randomly determined to be present according to the observed rates following either type of surgery defined by the systematic review, other literature source or expert opinion (Table 28). We assumed that there was no systematic co-occurrence of dysfunctions, so they were assigned independently. In this way it was possible for an individual to experience each dysfunction simultaneously.

FIGURE 19.

Schematic showing care pathway for individuals in long-term adverse event state with bladder neck contracture, urinary incontinence and erectile dysfunction. A3, adverse event state; B1, treatment of adverse event.

FIGURE 20.

Expanded care pathway for the management and treatment of individuals incurring long-term postoperative adverse events. A4, surveillance health state; B1, treatment of long-term adverse events.

TABLE 28 - Parameter values with distributions and sources for longer-term adverse events for individuals undergoing robotic or laparoscopic prostatectomy

MAPS, men after prostate surgery trial.

Upper and lower limits of triangular distribution calculated at ±25% of the point estimate. Upper and lower limits of log-normal distribution set at 95% CI.

Longer-term adverse event	Value	Probability	Lower limita	Upper limita	Assigned distribution	Source
Bladder neck contracture
Procedure rate robotic		0.008	0.002	0.052	Log-normal	Systematic review
Procedure rate laparoscopic		0.021	0.008	0.150	Log-normal	Systematic review
Urinary dysfunction management
Self-management < 1 year robotic		0.043	0.007	0.224	Log-normal	Systematic review
Self-management success at 1 year		0.957	0.720	1.000	Log-normal	MAPS cohort77
Self-management < 1 year laparoscopic		0.079	0.000	0.357	Log-normal	Systematic review
Surgical implantation of AUS	5.20%		3.90%	6.50%	Triangular	Clinical expert panel
AUS success rate	90.00%		67.50%	100.00%	Triangular	Clinical expert panel
Erectile dysfunction
Erectile dysfunction at 6 months	80.20%		60.00%	100.00%	Triangular	Stanford 2000187
Erectile dysfunction at 1 year	71.80%		54.00%	90.00%	Triangular	Stanford 2000187
Erectile dysfunction at 2 years	59.90%		45.00%	75.00%	Triangular	Stanford 2000187
Erectile dysfunction management
Treatment for erectile dysfunction	57.00%		42.80%	71.30%	Triangular	MAPS cohort (table 7.9)77
Reduction in erectile dysfunction treatment rate at 1 year	50.00%		37.50%	62.50%	Triangular	Matthew 2005188
Sildenafil: 100 mg once weekly	82.20%		61.70%	100.00%	Triangular	Schover 2002189
Sildenafil success rate overall	31.00%	0.690	23.30%	38.80%	Triangular	Blander 2000190
Alprostadil: 20 µg once weekly	15.40%		11.60%	19.30%	Triangular	Schover 2002189
Alprostadil success rate overall	57.10%	0.429	42.80%	71.40%	Triangular	Costabile 1998191
Penile prosthesis implantation	0.24%		0.20%	0.30%	Triangular	Schover 2002189
Penile prosthesis success rate	92.00%		69.00%	100.00%	Triangular	Meuleman 2003192

Perioperative costs

Costs associated with postoperative care

Extension of the time horizon to 40 years

In this sensitivity analysis we explored the impact of extending the time horizon. Conceptually this should allow more time for any benefits of robotic surgery to offset the increased procedure costs.

Changes in the costs of robotic equipment

Robotic equipment comes in several different variants and can be obtained from the manufacturers using several different payment plans. The precise cost of each of these variants may vary between provider and Appendix 12 provides illustrative examples of the cost variants. These costs have been converted into an annual cost, assuming the manufacturers’ recommended lifespan of the equipment of 7 years, and a cost per procedure estimated. In this analysis we explore what the impact on the incremental cost per QALY is of using a lower cost for the robotic system. This analysis has been repeated for the different numbers of annual cases performed (from 50 per year to 200 per year). From these results it was possible to determine the effect on estimates of cost-effectiveness of varying the cost per procedure of robotic prostatectomy consequent to any particular payment plan or throughput.

Changes in the risk of having a positive margin

The estimates of positive margin rates following robotic and laparoscopic surgery were based on the point estimates derived from the systematic review. In this sensitivity analysis we explored the impact of using both the lower and the upper 95% CrI limits of the OR of the difference in positive margin rates between robotic and laparoscopic surgery (base-case OR 0.69; 95% CrI 0.506 to 0.955). The further cancer treatment matrices defined by using the lower and higher risks of having a positive margin following robotic surgery are shown in Tables 36 and 37, respectively. The probabilities for laparoscopic surgery remained the same as in the base case.

Combining change in costs per procedure and positive margin rates

In this analysis we explored the impact on the incremental cost per QALY of changes in both the cost per procedure and the risk of a positive margin. These data have been presented as plots of the incremental cost per QALY against the positive margin rate, defined in terms of an OR, for different numbers of procedures performed per year.

Changes in the risk of biochemical recurrence

In the base-case analysis it was assumed that the risk of biochemical recurrence was the same regardless of which procedure a man received. The rationale behind this assumption was that the meta-analysis reported in Chapter 4 provided no evidence of any difference; the CI surrounding the OR was wide and included 1. In the first sensitivity analysis concerning biochemical recurrence rates we assumed that on average robotic surgery was associated with a lower rate of biochemical recurrence. This lower rate was estimated by combining the long-term rates from Menon and colleagues181 with the point estimate of the OR for risk of biochemical recurrence at 12 months obtained from the systematic review (0.89). The CIs around the OR were not clinically plausible and therefore we assumed a triangular distribution with upper and lower limits for the 12-month risk of biochemical recurrence for robotic surgery set at ±2% (based on the finding of Menon and colleagues181; Table 38).

In a second sensitivity analysis around the risk of biochemical recurrence we explored the impact of there being a higher rate of biochemical recurrence. The rationale behind this analysis was that the rates reported by Menon and colleagues181 were approximately 50% of those predicted in the meta-analysis. Therefore, in this sensitivity analysis we have simply doubled the rates observed by Menon and colleagues181 (Table 39).

Increasing the time horizon

When the time horizon increases, the costs and QALYs for both types of surgery increase; however, for all of the scenarios that were modelled (Table 41), costs increase only slightly whereas there is a much larger proportionate increase in QALYs. As a consequence, the incremental cost per QALY for all scenarios modelled is lower than in the base case and the probability of robotic surgery being cost-effective at threshold values for a QALY that society might be willing to pay197 increases towards 1.

Changes to the positive margin rate

In the base-case analysis we assumed that the OR for the difference in the positive margin rate between robotic and laparoscopic surgery was 0.69. In the first sensitivity analysis we took the difference in positive margin rates to be equal to the lower end of the CrI of the OR calculated in the meta-analysis reported in Chapter 4 (OR = 0.506). This resulted in robotic surgery having a lower rate of positive margins than in the base case and consequently a lower incremental cost per QALY (Table 42). Conversely, when the upper CrI limit of the OR for positive margins was used (OR = 0.955) the difference in positive margin rate between robotic and laparoscopic surgery was smaller than in the base case. As would be expected, the incremental cost per QALY increased as the number of procedures performed per year decreased. Indeed, only for the most optimistic scenario for robotic surgery modelled (the procurement cost of robotic equipment being equivalent to £2596) was the incremental cost per QALY < £30,000, and even in this scenario the likelihood that robotic surgery would be considered cost-effective was still only 60% at typical threshold values for society’s willingness to pay for a QALY (Table 43). 197 Overall, this sensitivity analysis illustrates the sensitivity of the results to changes in the effectiveness of robotic surgery because at the lower levels of throughput the mean incremental cost per QALY approaches or exceeds typical threshold values for society’s willingness to pay for a QALY (see Table 43). 197

Changes in the costs and positive margin rates

To explore the relationship between positive margin rates, incremental cost per QALY and cost per procedure, we have plotted the incremental cost per QALY for the different ORs for positive margin against the changing cost of the procedure determined by varying the number of procedures performed per year and the purchase cost of the robotic system (Figure 25). The data have been presented in this way as the cost per procedure is likely to vary markedly between centres according to throughput. The costs per procedure for different throughputs and for five alternative scenarios of robotic system cost are summarised in Table 44 (see Appendix 12 for details of how these costs were estimated).

FIGURE 25.

Incremental cost per QALY for different costs per procedure and relative differences in positive margin rates for robotic versus laparoscopic surgery. OR, OR for positive margin rate for robotic versus laparoscopic surgery.

As Figure 25 illustrates, as the cost per procedure increases with lower throughput and the OR for positive margin rate approaches 1 (no difference between procedures), the incremental cost per QALY increases beyond threshold values that society might be willing to pay. 197

For illustrative purposes these data have also been presented to show how the incremental cost per QALY changes as the relative difference in positive margin rate changes for different annual throughputs (Figure 26). As this figure illustrates, the incremental cost per QALY increases as the OR approaches 1.

FIGURE 26.

Incremental cost per QALY plotted against the OR for the relative difference in positive margin rate between robotic and laparoscopic surgery and for different numbers of procedures performed per year.

Changes to the risk of biochemical recurrence

In the base-case analysis it was assumed that the risk of biochemical recurrence was the same for both robotic and laparoscopic surgery. In the sensitivity analysis it has been assumed that on average robotic surgery is associated with a lower risk of biochemical recurrence (although the distribution attached to the value includes the possibility that there is no difference). A priori it would be expected that this would improve the relative efficiency of robotic surgery compared with laparoscopic surgery and, as Table 45 illustrates, on average this is what happened; however, the probability that robotic surgery would be considered cost-effective compared with the base case does not greatly alter over all threshold values considered.

In a second sensitivity analysis on biochemical recurrence rate we explored the impact of a higher risk of biochemical recurrence for both robotic and laparoscopic surgery (Table 46). The impact of this was to increase the costs of and reduce the QALYs from robotic surgery. As a consequence the incremental costs per QALY increased and for situations in which the annual number of procedures was ≤ 100 the incremental cost per QALY would be above thresholds currently adopted by NICE. 197 Consequently, the probability that robotic surgery would be considered cost-effective increases compared with the base case although at the lowest throughputs considered robotic surgery is still highly unlikely to be considered cost-effective (see Table 40).

Clinical effectiveness

The methodology used in this report makes best use of the current evidence comparing the safety and outcome of radical prostatectomy performed for men with localised prostate cancer by open, laparoscopic or robotic techniques. In the mixed-treatment meta-analysis, only studies that involved a comparator arm were included when estimating differences between treatments. It is noteworthy that none of the studies eligible for inclusion in the meta-analysis comes from a UK centre. The prevalence of radical prostatectomy for localised prostate cancer within a particular community or health-care system is predominantly governed by the prevalence of PSA testing, which continues to be low in the UK relative to other countries with similarly developed health-care systems. 35 Although we used uncontrolled data derived from studies performed in many different countries, we did not find any large discrepancies in demographic and disease variables that may have resulted in differences in outcome between UK men undergoing radical prostatectomy and those from other countries. In terms of the surgical teams, most will have undergone mentored training in established laparoscopic and robotic centres elsewhere in Europe or in the USA, with updates from conference and ‘master class’ attendance. Generalisation of our results to the UK context does seem appropriate given this face validity, but a degree of caution needs to be exercised.

As is commonly the case with attempts to summarise outcomes from treatments for prostate cancer, we were unable to identify comparative estimates of cancer survival. Instead, we had to use proxy measures of disease outcome including positive surgical margins and rates of biochemical recurrence at 1 year. 74 Although both are considered to be predictive of cancer-specific survival, proof of this relationship is lacking. 199,200 Despite these caveats, the findings from the systematic review on differences in the process of care, safety and cancer outcome between robotic and laparoscopic prostatectomy appear to have face validity. The systematic review involved > 19,000 men with an average age of 61 years with preoperative cancer characteristics that were balanced between the groups and consistent with current recommendations for the use of this treatment. 43 Overall, 96% of men had cT1–cT2 disease and 94% a Gleason sum score on preoperative biopsy of ≤ 7. Latest data from the British Association of Urological Surgeons (BAUS)201 on 2225 men undergoing radical prostatectomy, submitted by participating institutions in the UK during 2010, suggest that disease characteristics are similar in the UK, with a median age of 60 years, 92% having cT1 or cT2 disease and 93% a preoperative Gleason sum score of ≤ 7. Following surgery, the meta-analysis showed an overall upstaging, with 21% of men in both the laparoscopic and robotic groups being pT3, but no overall worsening of Gleason sum score. The proportion of men having pT3 disease is a key variable because it is predictive of both positive surgical margin rates and ultimate survival. Data from the 60 UK centres contributing to the BAUS 2010 dataset showed that 36% of men undergoing radical prostatectomy had pT3 disease. Additional recent case series from UK centres performing purely laparoscopic or robotic prostatectomy reported pT3 rates of 26% and 46% respectively. 156,177 In summary, men included in our study were broadly typical of the population requiring this intervention in the UK NHS, but with a possible lower rate of pT3 disease, reflecting higher use of on-demand PSA testing in the USA and other Western European countries.

Cost-effectiveness

No economic evaluations that compared the alternative forms of surgery from a UK perspective were identified and an economic evaluation based on a discrete-event simulation was planned. As described above, the findings of the systematic review were incorporated into the model and as a consequence the key determinants of cost-effectiveness were the time horizon, differences in positive margin rates and the relative costs of equipment. When a lifetime time horizon was adopted the costs and QALYs for both procedures increased but the increase in QALYs more than compensated for the increase in costs and hence the incremental cost per QALY was < £30,000 for all scenarios considered. This includes a scenario in which the number of procedures performed per year was 50 and in which the most costly robotic equipment was used. The principal reason for this is that adopting a longer time horizon allows more time for any benefits of robotic surgery to accrue and offset the initial higher equipment costs. Caution should, however, be exercised in interpreting the results as they rely on the extrapolation of relatively sparse short-term data within the model. There is uncertainty arising from both the quality of data and the mechanism for extrapolation.

The differences in positive margin rates translated into differences in QALYs and costs. For example, a higher positive margin rate resulted in lower QALYs, a greater need for further treatment and hence higher costs. With respect to costs, the cost per procedure was determined by the acquisition cost of the robotic system (which in turn depended on the specification of the equipment and the payment plan) and the number of procedures that might be performed annually using each robotic system. The costs of acquisition are to a certain degree under the control of a centre and depend on their own specific requirements and negotiations with the manufacturer. The number of procedures performed is a function of clinical need in the population that a centre serves and the population size. The results of the economic evaluation suggest that, when the difference in positive margins is equivalent to the point estimate estimated in the meta-analysis of all included studies, robotic radical prostatectomy was on average associated with an incremental cost per QALY that is less than threshold values typically adopted by the NHS when the cost of acquisition was low or the number of procedures was at the upper end of what could plausibly be achieved under current UK NHS provision (approaching 150 procedures per year). 197 This result holds except when the costs of acquisition were at the upper end of those estimated (see Appendix 12). Because the point estimate for difference in positive margin rate was uncertain, sensitivity analysis that progressively changed the difference in rates between robotic and laparoscopic prostatectomy was performed. At more optimistic values (OR = 0.506) the incremental cost per QALY would be less on average than threshold values typically adopted by the NHS when the number of procedures per year approached 100 or the procurement costs were at the lower end of those considered. Not unexpectedly, increasing the OR (OR = 0.955) resulted in a reduction in the QALY gain associated with the use of robotic prostatectomy and an increased cost. With the scenario of an OR for positive margin difference of 0.955 the incremental cost per QALY was only below the threshold if the number of procedures performed using each robotic system was increased to 200 and the lowest procurement cost for robotic equipment was assumed.

The mean estimates of incremental cost per QALY presented, although suggestive that robotic radical prostatectomy could potentially be cost-effective at conventional thresholds compared with laparoscopic prostatectomy, do not fully illustrate the degree of imprecision that exists. In the base-case robotic radical prostatectomy had an approximately 80% chance of being cost-effective when the threshold value for a QALY was £30,000. 197 However, caution should be exercised as this result does not incorporate the statistical imprecision surrounding variation in positive margin rates, a key predictor of longer-term outcomes in the model. This indicates the need for further data on the comparative long-term performance of the two forms of surgery. In addition, the sensitivity of estimates from cost-effectiveness for robotic prostatectomy to volume of surgery carried out in each centre argues for careful planning of NHS provision. As an illustration of the current service provision of the 60 UK centres that contributed to the BAUS radical prostatectomy database in 2010, 13 performed > 50 cases per year, of which three performed > 150 cases per year. 201 It should be noted, however, that less invasive management options for localised prostate cancer are emerging, including active surveillance, that may slow the growth in use of radical prostatectomy. 211

Clinical effectiveness

The strength of the study is the systematic approach taken to review the literature. Exhaustive systematic searches were made of the major electronic databases. All potential studies were reviewed for eligibility, including non-English-language publications. The risk of bias for each included study was assessed using the best available tool. To prevent biases caused by selective data extraction all outcome parameters were predetermined by expert panel consensus and any data were extracted using standard forms. Despite these efforts it is possible that some relevant data remained hidden as a result of non-publication.

In total, 54 primary comparative studies were included. Although this haul of relevant studies is impressive, not every study contributed data to each outcome. Furthermore, differences in reporting between studies also limited the opportunities for comprehensive meta-analysis. As a consequence of the limited evidence base, the CIs around many estimates of differences were wide and included differences that would be clinically important but could favour either treatment. Another major limitation resulted from the fact that the majority of comparisons were made against open radical prostatectomy, with few head-to-head comparisons of robotic and laparoscopic technologies. Thus, the estimates generated by the meta-analysis make use of indirect comparisons. The mixed-treatment comparison models used to handle such data are an effective method of handling evidence from many trials on several interventions in one analysis. 85 Like all analyses they require assumptions to be made that may or may not be reasonable and accordingly the results should be interpreted with a degree of caution. There were 80 non-randomised comparative studies in which the clinical stage of cancer at baseline was unclear, thereby excluding the studies from the review. Although every effort was made to contact the authors of those papers, only 19 replied. The subsequent finding that exclusion of 18 was appropriate provides some reassurance that these studies do not represent a source of missed useable data but there remains a possibility that some were excluded because of their inadequate reporting.

The review attempted to include only unique data from included studies but we experienced difficulty determining secondary publications because of a lack of clarity in reporting details of treatment centres. There were four study sets (Anastasiadis and colleagues122 and Salomon and colleagues;140 Ficarra and colleagues106 and Fracalanza and colleagues;107 Barocas and colleagues103 and Chan and colleagues;119 Greco and colleagues,129 Jurczok and colleagues131 and Fornara and colleagues127) in which details of the affiliated institute of the first author, type of treatment and treatment dates were similar but it was unclear from the reported text whether or not these studies included an overlap of the same men. It is therefore possible that five studies107,119,127,131,140 have contributed to an overinclusion of men for some perioperative and efficacy outcomes.

The risk of bias assessment in the conduct of a systematic review is important. For this review a robust combined checklist, developed by the Cochrane Collaboration Non-Randomised Studies Methods Group [Barnaby C, Reeves, Jonathan J, Deeks, Julian PT, Higgins, et al. on behalf of the Cochrane Non-Randomised Studies Methods Group. Chapter 13: Including non-randomized studies. In Higgins JPT, Green S, editors. Cochrane Handbook for Systematic Reviews of Interventions Version 5.1.0 (updated March 2011). The Cochrane Collaboration, 2011. URL: www.cochrane-handbook.org (accessed March 2011)], assessing different sources of bias was produced. A scoring scale approach based on design features was avoided as this has been reported to be inaccurate concerning the direction of bias and can include items that are unrelated to the internal validity of a study. 212 For example, the terms ‘prospective study’ and ‘retrospective study’ are particularly ambiguous. ‘Prospective study’ should imply that all design aspects were planned, including hypothesis generation, recruitment of participants, baseline data collection and outcome data collection. In practice, how prospective a study is can often be unclear as some aspects of a study can be prospective, such as hypothesis generation and determination of outcomes, whereas others are retrospective, such as length of stay data collection from hospital records. The potential for bias in designs with different attributes can therefore vary considerably. This systematic review identified few studies at low risk of bias. The moderate inter-rater agreement between the two independent reviewers that was found in our review illustrates that risk of bias can be interpreted in different ways by different people. This is particularly likely in the newly developing methodological area of summarising non-randomised studies in which the level of reporting is often poor.

Many studies failed to report point estimates and measures of variability, hindering their use in estimating weighted mean differences, which require mean estimates for each intervention and standard deviations. It is possible that if means and standard deviations were reported more consistently, effect sizes would be different. However, in the systematic review, when an appropriate measure of variability was not reported for continuous outcomes, consistency across studies reporting the outcome was investigated and this would serve to eliminate biases when determining the direction of effect, even though the magnitude of effect remains uncertain.

A more specific methodological limitation that frustrated pooled analysis was the use of differing definitions and measures of functional outcomes for both urinary and erectile dysfunction. The variety of different ways of measuring dysfunction reduced the ability to compare data or to conduct a comprehensive meta-analysis. This was in part reflected by changing measurement methodologies for dysfunction across the time frame over which the studies were conducted, but it will remain a problem until consensus on important outcome measurements in this clinical area can be agreed. Initiatives such as the UK Medical Research Council-funded Core Outcome Measures in Effectiveness Trials (COMET) initiative213 may be useful in this context. Such initiatives aim to help researchers and clinicians across all specialities to develop a standardised set of outcomes (or core outcomes) that should be measured and reported as a minimum in all clinical trials of a specific condition, in order to make it easier to compare, contrast and synthesise the results of trials, to reduce the risk of inappropriate outcomes being measured and to reduce outcome reporting bias. 214

The examination of the influence of learning curves on the results was limited by poor reporting in the included studies. Given the general lack of data reported on the experiences of the centres included in the review, a proxy measure of ‘experience’ was used – namely the number of procedures performed. This measure may be inadequate to detect the differences between the interventions. In addition, when learning curve data were obtained from case series, the reported improvement with increasing experience may have limited applicability to current practice. This is partly because of the early reports of the effects of laparoscopic procedures focusing on refining the technique rather than on the acquisition of the technical skills required to perform the procedure in routine practice. If future studies conform to CONSORT reporting standards for non-pharmaceutical interventions215 this may help to alleviate some of the problems.

In summary, we believe that we have used the best available techniques to identify, review and meta-analyse the data that were available to us. This approach has enabled us to make robust broad conclusions concerning the relative beneficial and adverse effects of robotic prostatectomy compared with laparoscopic prostatectomy but which are associated with a defined degree of uncertainty.

Discrete-event model and economic evaluation

The economic evaluation was based on a discrete-event model. The purpose of this model was not just to estimate relative cost-effectiveness but also to investigate potential differences in clinical outcome between laparoscopic and robotic radical prostatectomy. As the model is a further level of evidence synthesis that builds on the systematic review and meta-analysis, many of the limitations applicable to the clinical data also apply to the economic data.

The decision context, like many of those faced in the evaluation of health-care interventions, was complex. Within a clinical context there is considerable variation between individuals in terms of demographic status and disease progression. In addition, the range, frequency and management of postoperative adverse events following surgery and the variations required in the care pathways necessitated the use of a more complex model than originally envisaged. The model form adopted was able to incorporate the degree of heterogeneity needed to simulate the life trajectory of individuals following surgery. In developing this model, we did not compromise realism in defining how care was implemented in the model. Elements of care that could occur in a given clinical setting were included insofar as they were recognised by the expert panel of practitioners. This inclusive approach effectively led to a complex suite of pathways that could not be modelled using ‘off-the shelf’ modelling packages often used in economic evaluations.

The complexity of the model permitted the simulation of a multitude of possible patient trajectories through the model. This can be illustrated by taking the example of a man who presents with a tumour of stage cT1 and undergoes surgery for presumed localised cancer. On pathological examination of the removed prostate it might be found that the tumour margin is positive but he is counselled to continue under surveillance with regular PSA checks. Happily there is no sign of biochemical recurrence and he remains in the surveillance state until the end of the 10-year time horizon of the study. In a more complex case, a man might remain under surveillance without cancer recurrence but require treatment for urinary dysfunction; he then subsequently requires further treatment for a localised recurrence, which is unfortunately unsuccessful, and he dies of prostate cancer following a period on androgen deprivation therapy. These complexities are required to model the costs and consequences of the differential outcomes of clinical effectiveness found in the systematic review but have the disadvantage of increasing the potential for error and misattribution. To guard against this the longer-term outputs of the model were checked for plausibility and credibility against existing literature sources and the opinions of our expert panel.

The major drivers of model design were heterogeneity in disease status and the requirement to describe realistic care pathways reflecting the range of postoperative adverse events and their treatment. Each health event and postoperative change in management was modelled probabilistically based on available data. As described in Chapter 5 this involved first defining the risk of an event occurring and then, for each man in a simulated cohort, generating a random number between 0 and 1. If the random number was less than the defined risk then the event was assumed to have occurred for that man. This process inevitably led to a large data requirement and a trade-off between model accuracy and data availability.

The data used within the model came from a number of, often independent, sources, which ranged from quantitative data derived from the systematic review through to qualitative data provided by clinical expert members of our advisory panel. Furthermore, parameter estimates for each event were assumed to be unbiased and representative of the population of men requiring radical prostatectomy for localised prostate cancer in the UK NHS. The use of different data sources, although unavoidable, may have introduced biases into the model estimates as the data came from different samples of the worldwide population of men undergoing radical prostatectomy. Furthermore, it was not always possible to assess the likelihood of non-independence in the parameter estimates. To overcome these limitations the parameters estimates were validated by the expert panel and model output discussed within the project team for clinical plausibility.

To address the imprecision we incorporated estimates of uncertainty for some parameters from the results of the meta-analysis. For other parameters we assumed triangular distributions when we had some information on mid-point and upper and lower limits for parameters and then used sensitivity analysis to investigate the behaviour of the model when we varied parameters for which we had only a point estimate and which were crucial to the model output. The sensitivity of health-related and economic outcomes was explored by determining the impact of varying the two parameters perceived to be of crucial importance to overall outcome: rates of pathological positive margin status and incidence of biochemical recurrence. In the case of positive margin rates the parameter was only one of the inputs used for deciding the need for further cancer treatment postoperatively. This precluded the exploration of imprecision in the probabilistic analysis and therefore this parameter was the focus of extensive deterministic sensitivity analysis.

When considering the impacts of each intervention strategy on health states, further treatment for cancer following radical prostatectomy was estimated as a less frequent event following robotic surgery than following laparoscopic surgery. This resulted in fewer cancer-specific deaths following robotic radical prostatectomy than following laparoscopic radical prostatectomy. The consequence of this was greater QALYs following robotic surgery and it also partly compensated for the increased costs of the robotic equipment.

Despite considerable efforts to elicit relevant information it was not possible to precisely quantify the extra cost of the robotic surgery equipment per procedure. This was because there are a plethora of different procurement strategies provided by the manufacturer, Intuitive Surgical, which varied by both method of payment and specification of equipment. Furthermore, the number of procedures performed each period using a given piece of equipment is variable. In the base case we chose to use the highest procurement cost and the highest plausible throughput of 200 cases per year. Repeating the analysis using lower procurement costs and a reduced number of procedures resulted in variation in the proportion of the cost of the robotic system attributed to each procedure, from £3500 to £10,200 (see Table 40). In the base-case analysis, only when the cost was at the higher level determined by a throughput of approximately 150 cases per year was the incremental cost per QALY around £30,000. It should be noted that more favourable assumptions around the positive margin rate tended to reduce the incremental cost per QALY but the incremental cost per QALY would still be > £30,000 for annual throughputs of approximately 100 cases (or a cost of robotic equipment per procedure of approximately £6000). It should also be noted that less favourable but still plausible assumptions concerning the difference in positive margin rates also increased the incremental cost per QALY to > £30,000, particularly when combined with lower throughput of cases. These results indicate that further research is required to more accurately determine positive margin rates and also how they predict long-term cancer outcomes.

In addition to clinical data and costs the model also attempted to incorporate information on the value of different events to the men under treatment – health-state utilities – so that QALYs could be estimated. Searches were conducted to identify data of most relevance to a UK decision-making context but few data were found and not all data were available from a single source. It is possible that we may have misvalued some events, which, if these events occurred at different rates between the two procedures, would have introduced a bias into the analysis. Ideally, health-state utilities data applicable to a UK population should be elicited to overcome this shortcoming.

One aspect of cost not included in the model was the use of unscheduled GP and outpatient visits. There was a lack of data on the frequency of these events with which to model. Previous experience from trials that include men after treatment of prostate cancer would suggest that these costs are relatively modest compared with the cost of surgery. Furthermore, given the apparent lack of difference in effects we did not expect there to be a substantial differential use of these services between groups.

In summary, the discrete-event model attempted to synthesise current clinical practice with the best available estimates of economic and health data to evaluate the potential benefits of robotic prostatectomy in comparison with standard laparoscopic prostatectomy. The model was conservative in that we did not model processes for which we had no evidence of a difference between the two surgical approaches. Furthermore, it did not assume dependence between processes when there was no information available to support a modelled relationship. The model demonstrated that there are circumstances when robotic prostatectomy could be cost-effective as judged against conventional thresholds for willingness to pay for a QALY, especially if lower costs of equipment can be secured and when the surgical capacity is high.