Transperineal biopsy devices in people with suspected prostate cancer - a systematic review and economic evaluation

Epidemiology

In 2018, there were 49,810 new diagnoses of prostate cancer in England, an increase of 7985 more registrations than the previous year. 5 The age-standardised incidence rate in England was 204.7 per 100,000 in 2018, which was an increase from 182.8 per 100,000 in 2009. 6 The incidence rate for prostate cancer in the UK is projected to rise to 233 cases per 100,000 males by 2035. 1

Prostate cancer accounts for 30% of all male cancer diagnoses and is the most commonly diagnosed cancer in males over 45 years old. In 2018, 55% of prostate cancers were diagnosed at stages 1–25 and despite an increased incidence rate the age-standardised mortality rate decreased between 2009 and 2018 from 51 per 100,000 to 46 per 100,000. 6

In England, the South East has the highest age–sex-standardised rate of prostate cancer (228 per 100,000 people), compared with the North West at 171 per 100,000 people. 5 Prostate cancer incidence rates in males in England are 17% lower in the most deprived quintile compared with the least deprived quintile (2013–7). 1 Cancer Research UK states that ‘Prostate cancer is most common in black males, then white males and least common in Asian males’. 1

Transrectal ultrasound prostate biopsy

During a transrectal ultrasound (TRUS) prostate biopsy a TRUS probe is inserted into the anus to image the prostate. Samples of prostate tissue are collected using a biopsy needle inserted via the anus, through the rectal wall, and into the prostate. This procedure is typically carried out under local anaesthetic in an outpatient setting but can also be carried out under general anaesthetic (e.g. if the patient is unlikely to be able to tolerate the procedure under local anaesthetic). However, because the biopsy needle is inserted through the rectal wall, biopsy-related infections can occur, including, in some cases, sepsis (estimated to be 0.8% in a 2016 systematic review). 14 Sepsis is a serious infection which requires a hospital admission and antibiotics.

Traditionally, most prostate biopsies in the NHS used the TRUS method. However, there has been an increase in the use of transperineal biopsy (TP), and this has been accelerated due to the COVID-19 pandemic. A strategy document issued by the British Association of Urological Surgeons (BAUS) Section of Oncology for the interim management of prostate cancer during the pandemic recommended that TRUS biopsies should be avoided if possible. 15

Transperineal prostate biopsy

In common with TRUS, a transperineal prostate biopsy also uses a TRUS probe inserted into the anus to image the prostate, but the samples of prostate tissue are collected using a biopsy needle inserted through the perineum (the skin area between the anus and the scrotum) rather than through the rectal wall. Transperineal prostate biopsy can be conducted using any of the following methods:

• a grid and stepping unit

• a coaxial needle (‘double freehand’)

• a freehand device (using one of the six devices listed in the NICE scope for this assessment).

Clinically significant prostate cancer

When prostate cancer is diagnosed, it is often distinguished in terms of whether the cancer is CS or insignificant. The purpose is to assess how rapidly the cancer will progress and, hence, whether to recommend active surveillance or active treatment. Expert clinical opinion suggests there is no universally agreed definition of the term CS prostate cancer. There are varying definitions available in the literature. For example, clinicians at University College London (UCL) devised criteria for defining CS cancer, as localised cancer with a maximum total cancer core length of 10 mm, a maximum cancer core length of 6 mm and a Gleason score of at least 4 + 3 or 3 + 5 (UCL definition 1). A second set of criteria from this group defines CS cancer as a maximum total cancer core length of 6 mm, a maximum cancer core length of 4 mm and a Gleason score of at least 3 + 4 (UCL definition 2). These criteria have been used in clinical trials assessing different prostate biopsy modalities, including the PROMIS trial in the UK, which examined the diagnostic accuracy of mpMRI and TRUS biopsy in prostate cancer. 17

The NICE clinical guideline prostate cancer diagnosis and management (NG131) defines CS prostate cancer as any prostate cancer of Gleason score 7 and above. 18

Population

The relevant population is people with suspected prostate cancer where prostate biopsy is indicated. People included in the review may have a clinical suspicion of prostate cancer (e.g. raised PSA level or abnormal DRE findings), or people may have had a previous prostate biopsy that was negative for prostate cancer but have a continued clinical suspicion. People are not included if they have already been diagnosed with prostate cancer and are receiving treatment or monitoring by active surveillance or by watchful waiting, and likewise people are not included if they are known to have metastatic prostate cancer.

Interventions and comparators

Local anaesthetic transperineal prostate biopsy is the diagnostic procedure relevant to this review, and for the purposes of this report is considered as the intervention. The relevant LATP procedures vary according to two separate (though related) decision questions.

• Decision question 1 compares any LATP prostate biopsy procedure versus LATRUS prostate biopsy or versus GATP prostate biopsy. For example:

  • LATP using a grid and stepping unit

  • LATP using a coaxial needle (‘double freehand’)

  • LATP using a freehand TP device (see decision question 2).

The comparison of LATP versus LATRUS assess differences/similarities in diagnostic and clinical outcomes between the transperineal and transrectal prostate biopsy respectively, both using local anaesthetic. The comparison of LATP versus GATP assesses differences or similarities in diagnostic and clinical outcomes between different anaesthetic modalities used during the transperineal prostate biopsy.

• Decision question 2 compares LATP using any of the six freehand devices listed below versus LATRUS, GATP or LATP using a grid and stepping unit (NB: name of the company making/distributing the device in parentheses):

  • PrecisionPoint (BXTAccelyon)

  • UA1232 (BK Medical)

  • Trinity Perine (KOELIS/Kebomed)

  • CamPROBE (JEB)

  • SureFire Guide (LeapMed)

  • EZU-PA3U (Hitachi).

As evident from the above, the intervention relevant to decision question 2 (LATP using any of the six freehand devices) is nested within the broader range of biopsy interventions relevant to decision question 1 (any LATP prostate biopsy procedure). The comparators relevant to decision question 2 overlap with those relevant to decision question 1, but additionally, include LATP using a grid and stepping device (see Table 1 for a summary of the above).

No restriction was placed on the inclusion of specific biopsy protocols and procedures, such as number of biopsy cores taken, or whether prostate biopsy sampling was systematic and/or targeted, and whether mpMRI was used to determine whether a prostate biopsy is needed, and, if so, which prostate lesions should be targeted for core sampling. Cognitive fusion biopsies, also known as visual registration biopsies, were eligible, whereas software-based fusion biopsies were not. Biopsy techniques using sedation in place of local or general anaesthetic were not included.

Outcomes

We categorised relevant outcome measures according to which aspect of the prostate biopsy they evaluate, following the same approach used in the NICE scope for this diagnostic assessment. Our synthesis of the results of the studies is structured according to these categories for consistency and ease of report navigation (see Intermediate outcomes, Clinical outcomes and Patient-reported outcomes).

Intermediate and diagnostic outcomes of relevance were: measures of diagnostic accuracy (e.g. sensitivity/specificity); cancer detection rates; CS cancer detection rates; clinically insignificant cancer detection rates; low-, medium-, high-risk cancer detection rates; biopsy sample suitability/quality; number of biopsy samples taken; procedure completion rates; re-biopsy events within 6 months and length of time to perform the biopsy procedure (we added the latter outcome to inform biopsy cost estimates for potential inclusion in our economic model to assess cost-effectiveness; see Economic analysis).

Clinical effectiveness outcomes of relevance were hospitalisation events after biopsy; rates of biopsy-related complications, including infection, sepsis and haematuria; rates of urinary retention; rates of erectile dysfunction; survival; progression-free survival; adverse events from treatment.

Patient-reported outcomes of relevance were HRQoL and patient-reported tolerability. We added biopsy procedure time to the inclusion criteria for outcomes because it impacts on the cost of the procedure.

Study design

Any primary comparative research study evaluating the biopsy methods outlined in the ‘Interventions and comparators’ subheading above is included. We noted single-arm evaluations of LATP biopsy during screening so that we could potentially include them if there was insufficient available comparative evidence.

Overview of general study characteristics

Table 3 gives an overview of the LATP prostate biopsy versus LATRUS biopsy studies included in the review.

Of the 15 included studies comparing LATP-any versus LATRUS biopsies, 5 are RCTs, 7 prospective cohort studies and 3 retrospective cohort studies.

The RCTs were conducted in Japan,26,28 China,25 Hong Kong27 and Italy,24 and all were single-centre studies. The participants in all RCTs were prostate biopsy naïve with suspected prostate cancer, and no study reported any pre-biopsy mpMRI. The LATP techniques varied: one study used a coaxial needle,24 another used an unnamed attachment for needle guidance,28 another used PrecisionPoint27 and two studies did not specify a device. 25,26

The seven prospective cohort studies are all single-centre studies, set in England,29,34–36 Hong Kong,33 Japan,38 Singapore30,31 and Italy. 32 They comprise two studies which carried out both transperineal and transrectal biopsies in the same participants in the same session,32,38 three studies where the LATRUS arm is a historical comparison group,29–31,34,35 one study that assigned participants to study arms according to pre-biopsy MRI findings and other criteria,36,37 and one study that does not report how it assigned participants to study arms. 33

The participants in the three English prospective cohort studies are a mixed population of those who were biopsy naïve, those who were undergoing repeat biopsy and a small proportion of participants on active surveillance. In all the other studies participants were exclusively prostate biopsy naïve. All English studies used the PrecisionPoint device to perform LATP,29,34–37 as did the Hong Kong study,33 and the earlier studies do not report any device. 32,38

One of the studies33 is reported only in a conference abstract and another is an unpublished slide-set presentation29 and so they have limited information. The other studies are reported in full publications.

The retrospective studies were set in Italy,39 China40 and the USA. 41 The Italian and Chinese studies were multicentre (two-centre) studies where LATP was performed at one centre and LATRUS was performed at the other. The USA study is a single-centre study. One study population consists entirely of repeat biopsy participants39 one study consists entirely of biopsy naïve participants,40 and one study included a mixed population of biopsy naïve, repeat biopsy and active surveillance participants. 41 Two studies performed propensity score matching of the participants: one study reports propensity score matched results only39 and the other reports both the unmatched and propensity score matched results. 40 The LATP techniques varied according to device used: one study used a coaxial needle,39 one study used the PrecisionPoint freehand device41 and one study did not report using a device. 40

Details of local anaesthetic transperineal prostate-any biopsy procedures

Table 53 in Appendix 4 gives details of the LATP-any biopsy procedures. Most studies used systematic biopsy sampling, with the number of cores taken (where reported) ranging from 6 to 24 across studies. Two studies based the number of cores taken on the size of the prostate, one by whether or not the prostate volume was above or below 50 ml,25 and another study reports that the samples were spaced 1 cm apart. 41

Where targeted biopsy sampling was performed this could be in addition to systematic sampling biopsies, or targeted sampling alone. 34,35 Reasons to prompt additional targeted sampling were: suspicious areas detected by TRUS or DRE,25 any hypoechoic areas noted,32 PI-RADS score > 2 on pre-biopsy mpMRI,36,37 hypoechoic lesions or palpable nodules on DRE,38 or participants with pre-biopsy mpMRI PI-RADS score of 4 or 5. 36

Additional variations to the biopsy procedures that are not reported above are: any other medications administered or ceased (e.g. anticoagulation medication), whether antibiotic prophylaxis was given (and how much), what position the participant was in (e.g. lithotomy or dorsal lateral) and where they were performed (e.g. in outpatient clinics or day theatres), thus further illustrating the heterogeneous nature of the biopsy procedures and the studies.

Participant characteristics

Most of the included studies reported age, PSA level, prostate volume and the proportion of participants with abnormal DRE or pre-biopsy imaging findings (see Table 54, Appendix 4).

Age is reported in various combinations of mean or median with interquartile range (IQR), range or standard deviation (SD), with a mean age of 63–72 years across the studies. PSA level is also reported in various combinations of mean or median with IQR, range or SD. It can be seen that mean PSA levels varied from around 7–8 to 12–19 ml, with one of the retrospective studies having participants with PSA levels 38–40 (Jiang et al.). 40 Only five studies reported PSA density. 28–31,34,35,41

The PI-RADS score, based on pre-biopsy imaging, is only reported in two studies neither of which correspond exactly with the NICE subgroups of interest (people with a Likert or PI-RADS score of 2 or less, or a score of 3, 4 or 5). One study reports the proportion of participants with PI-RADS 2/3, 3/4 and 5 separately, but only for the LATP arm. 34,35 The other reports the proportion of participants with PI-RADS 4 or 5. 41 None reported the location of lesions identified in pre-biopsy imaging.

Two studies reported body mass index,24,25 one study reported ethnicity. 41 None reported any family history of prostate cancer.

There is not enough evidence to review the efficacy of the biopsy procedures for several of the NICE subgroups (people with anterior lesions; people with posterior lesions; people with apical lesions; people with basal lesions; people with a Likert or PI-RADS score of 2 or less; people with a Likert or PI-RADS score of 3, 4 or 5).

Summary

The comparison of LATP-any versus LATRUS biopsy (decision question 1) is the largest in terms of number of included studies, comprising five RCTs, seven non-randomised prospective studies and three retrospective studies. This is not unsurprising given the broad scope of the LATP-any intervention grouping in this assessment, which encapsulates the spectrum of transperineal prostate biopsy techniques in use. Three studies (non-randomised) were set in England, but many were done in East Asian countries. The vast majority of study participants were prostate biopsy naïve with suspected prostate cancer, with just one study assessing the effects of repeat biopsies in people with suspected prostate cancer who had a previous negative biopsy. The TP biopsy protocols (e.g. device used/sampling method/number of cores taken) varied between studies, which may partly reflect local clinical practice guidelines in study host institutions, but also the evolution of transperineal prostate biopsy practices over time (e.g. increases in the number of cores sampled). Some of the more recently published studies used pre-biopsy mpMRI to inform biopsy sampling, but this constitutes a small proportion of the evidence base as a whole.

Overview of general study characteristics

Table 4 gives an overview of the four studies comparing LATP-any biopsy versus GATP biopsy with grid and stepping device. Three of the studies are available only as conference abstracts currently; thus some of the necessary detail in the following subsections is limited. 43–45

Of the four studies, one was a RCT set in China,42 two were prospective non-randomised studies set in England44 and Japan43 respectively, while the fourth was a retrospective study set in New Zealand. 45

One study42 used a grid and stepping device to perform LATP biopsy; another performed LATP using the PrecisionPoint freehand device45 and two studies did not specify use of a device. 43,44

Details of prior biopsy history were not clearly reported, but in one study it is stated that all participants had previously had one or more negative biopsies. 43

Details of local anaesthetic transperineal-any biopsy procedures

Table 55 in Appendix 4 gives details of LATP-any biopsy procedures used. Reporting of details by the studies was limited, but the available information shows that systematic sampling was commonly performed, with additional targeting of cores based on pre-biopsy imaging. Details of image guidance and anaesthesia are limited.

Participant characteristics

Available information on the characteristics of study participants (e.g. age, PSA level, prostate volume) is extremely limited, and only one study gave adequate detail (see Table 56, Appendix 4). 42

Summary

This comparison (LATP vs. GATP, decision question 1) is based on a smaller evidence base: one RCT, two prospective observational studies and one retrospective observational study. The location of the studies is mixed, including two studies done in Asia, and one each from New Zealand and England, respectively. LATP was performed using a grid and stepping device in at least one study, and using a freehand device (PrecisionPoint) in another. Sampling was systematic with additional targeting of cores in some cases. With the exception of the RCT, the other three studies are reported in conference abstracts only, thus limited information is available.

Overview of general study characteristics

Seven studies were identified that compare LATP biopsy using a freehand device with LATRUS biopsy. All freehand devices are the PrecisionPoint device; see Table 10. In contrast, only one study compares LATP biopsy using a specific freehand device with GATP (n = 1, PrecisionPoint device); see Table 12. No studies were identified that compare LATP-freehand with LATP using a grid and stepping device.

As no comparative studies were identified for any devices other than PrecisionPoint, we included single-arm studies for devices where no comparative evidence was available. One study reports a single-cohort study (i.e. with no comparative biopsy group) reporting ‘the first in man’ evaluation of the CamPROBE device. 46 Three conference abstracts report three separate single-cohort studies that used the UA1232 device. 47–49 See Table 13.

Of the seven studies comparing LATP-PrecisionPoint to LATRUS, one is a RCT,27 five were prospective cohorts29–31,33–37 and one was a retrospective case series. 41 All studies were single-centre studies, with three conducted in the England, two in Hong Kong, one in Singapore and one in the USA. The English and American studies were of mixed populations, whereas the others were prostate biopsy naïve participants with suspected prostate cancer only, and only two studies reported the number of cores taken during biopsy: 12 cores30 and 24 cores. 29

Participant characteristics

Participant characteristics are reported for the LATP freehand device PrecisionPoint versus LATRUS studies and are summarised in Table 57 in Appendix 4.

Summary

The evidence for this comparison (LATP-freehand vs. LATRUS, decision question 2) is a subset of the evidence for the LATP-any versus LATRUS, decision question 1 comparison. All the evidence is for the PrecisionPoint freehand device as the intervention. Included within this set of seven studies is one RCT and the three non-randomised studies set in England.

Overview of general study characteristics

Table 6 gives an overview of the single study comparing LATP-PrecisionPoint versus GATP biopsy. 45

Rij and Chapman report a retrospective cohort study conducted in a single centre in New Zealand. 45 At the current time (November 2021) the study is available publicly only as a conference abstract. The precise details of the study methods and outcomes are therefore limited. This study did not report the indications for biopsy, nor the number of cores taken during the biopsies, nor any participant characteristics.

Overview of general study characteristics

No comparative evidence was identified for the LATP freehand devices CamPROBE, UA1232, SureFire, EZU-PA3U and Trinity Perine Grid. We therefore looked for any relevant single-arm (non-comparative) studies of these freehand devices. We did not identify any relevant single-arm studies with SureFire, the Trinity Perine Grid (for which all the studies we found used software-based fusion techniques outside the scope of this review) or EZU-PA3U. Table 7 gives an overview of the CamPROBE and UA1232 studies.

The one study evaluating CamPROBE was a prospective single-cohort study (i.e. with no comparative biopsy group) conducted in six centres in England. 46 It has a small (n = 40) study population. The indications for prostate biopsy were not reported and two devices were used per patient per biopsy, for the right and left sides of the prostate, respectively.

The three studies evaluating the UA1232 device are all single-centre prospective single-cohort studies conducted in England. 47–49 The study populations are larger (n = 482, n = 200, n = 219) and all the participants are biopsy naïve. All three studies were identified via the company submission as none of the abstracts explicitly report using the UA1232 device. All are conference abstracts and as such contain limited information.

Participant characteristics

The reporting of participant characteristics for the single-arm studies for CamPROBE and UA1232 is minimal: the CamPROBE study46 reports participants’ median and range for age, and one of the UA1232 studies47 reports median age and median PSA level.

Summary

The evidence available for LATP-freehand devices specified in the NICE scope, other than the PrecisionPoint device, is limited to single-arm studies: CamPROBE46 with a small population and UA1232 with limited information from three conference abstracts. 47–49 There is no evidence for the other devices in the NICE scope. Details of study characteristics and participant characteristics are limited.

Critical appraisal of randomised controlled trials

As mentioned earlier (see Critical appraisal of study methodology), we used the Cochrane risk of bias tool (version 1)20 to critically appraise the six RCTs in our review. 24–28,42 A key finding from this exercise is that we are unable to fully judge the studies’ overall risk of bias due to inadequate reporting of study methodological details in the available publications. Commonly, therefore, we recorded ‘unclear’ risk of bias for studies across the domains, notably those concerning reporting bias (due to selective outcome reporting), detection bias (due to lack of blinding of outcome assessors to type of prostate biopsy performed) and selection bias (due to inadequate randomisation of participants to trial arms, and/or inadequate concealment of the randomisation sequence). However, sufficient detail was available to inform judgements relating to other bias domains, including attrition bias. Overall, we advise caution in the interpretation of these study findings due to uncertainty regarding potential risks to their internal validity. Below is a brief summary of our findings; full details are reported in Table 59 in Appendix 5.

There was a lack of detail given on the methods used for random sequence generation in four of the trials,24,26–28 leading to uncertainty about whether or not ‘true’ randomisation had been achieved and selection bias avoided. Likewise, little or no information was given on whether adequate procedures were in place to conceal the random allocation sequence from study personnel, particularly those involved in enrolling participants to the study.

We judged all six trials to be at high risk of performance bias on the reasonable assumption that study participants and investigators knew which type of biopsy procedure participants had been randomly allocated to. This is an unavoidable consequence of this type of intervention, whereby the clinician performing the biopsy cannot be blinded to the type of biopsy the participant has been allocated to. Likewise, it is unlikely that the study participant would not be informed of their surgical procedure. It is also unclear whether any protocols were in place to reduce the risk of differential behaviours by participants and healthcare providers associated with knowledge of the type of biopsy performed. All six trials were judged at low risk of attrition bias, due to no or minimal reported participant loss to follow-up or study withdrawal.

Our judgements of the risk of bias across the five domains were identical for four of the six RCTs. 24,26–28 The trial by Guo et al. was at low risk of bias for the greatest number of domains:25 specifically, low risk of detection bias due to blinding of the outcome assessor (pathologist), low risk of selection bias due to adequate (computer-generated) randomisation (though we cannot rule out selection bias completely because details of allocation concealment were not reported) and low risk of reporting bias.

Critical appraisal of observational studies

As stated earlier (see Critical appraisal of study methodology) we used the checklists from the JBI suite of critical appraisal tools to critically appraise observational studies. 21 Eleven of the 13 observational studies were assessed using the JBI checklist for cohort studies50 and the remaining 2 studies were assessed using the JBI checklist for case series. 51

Most of the cohort studies recruited biopsy comparison groups from the same or a similar population. Likewise, the case series reported consecutive/complete inclusion of participants. However, limited reporting of study inclusion criteria and participants’ demographic and clinical information means it is unclear how comparable the biopsy groups within the studies are. Confounding factors were identified and handled in only about half of all the studies (both cohort studies and case series); the remainder are mostly unclear. Therefore, we judge the studies to have unclear risk of selection bias.

Follow-up times and methods to deal with loss to follow-up were mostly unclear, raising the potential for attrition bias. However, some key outcomes relevant to this diagnostic assessment are unlikely to be affected by loss to follow-up as they are measured/taken during the biopsy procedure itself (e.g. cancer detection rate based on biopsy samples) or immediately afterwards (e.g. pain questionnaires). Therefore, we judge the risk of attrition bias as low for cancer detection rate and pain/tolerability outcomes, but unclear for other outcomes.

The risk of detection bias was judged as generally low because in almost all the studies the biopsy methods are clearly reported and over half of the studies reported using a protocol or schema for the biopsy procedure. In addition, the cancer detection rate outcome was measured in a valid and reliable way in most of the studies, usually referring to a specific grade group or score. However, there may be a risk of detection bias when considering the validity and reliability of measurement of the other outcomes in several of the studies, for example complications, where for some studies only complications that occurred were reported and no time frame was stated for reporting any complications. Therefore, when considering different outcomes in the studies, detection bias is either low or unclear depending on the outcome in question (as for attrition bias).

There is a high risk of reporting bias (and several other bias domains) in studies available, at the time of writing, only as conference abstracts. Commonly, abstracts are restricted in word limits, prohibiting authors from reporting all intended outcome data. Clarity on reporting bias may improve if full text reports of studies are published (personal communication with study authors indicates that some are in the process of preparing manuscripts for publication). There is lack of clarity around several domains of bias due to the limited amount of information that can be conveyed in a conference abstract.

Our critical appraisal judgements for each cohort study and each case series are presented in Table 60 in Appendix 5.

Prostate cancer detection (local anaesthetic transperineal-any biopsy vs. local anaesthetic transrectal ultrasound, decision question 1)

Prostate cancer detection was the most commonly reported of all the outcome measures relevant to this assessment (n = 14 of 15 studies). Only the study by Starmer et al. did not report this outcome. 36,37 In marked contrast, CS prostate cancer detection, informative for assessing the risk of rapid cancer progression, was reported in just five studies. 27,29,33–35,41 Table 8 reports study cancer detection rates, including CS cancer rates, where available.

There was variation between the studies in overall cancer detection rates, which highlights the heterogeneous nature of this evidence base. In terms of differences in detection rates between LATP and LATRUS, the results are mixed. Some studies reported similar rates for the two biopsy methods, while others reported differences in rates. There is not a clear pattern to these differences – in some cases LATP biopsy detects a greater proportion of cancers than LATRUS, but the opposite is also evident. We urge caution when interpreting these results given the predominance of observational study methods. The similarities and differences in cancer detection rates between the two biopsy methods may be driven, in part, by selection bias from lack of random allocation of participants to LATP biopsy or LATRUS biopsy study arms.

Figure 9 in Appendix 4 shows pooled study estimates from a random-effects meta-analysis of LATP-any versus LATRUS for detection of prostate cancer. There is no statistically significant difference between LATP-any biopsy and LATRUS biopsy in detection of prostate cancer based on RCTs (RR = 1, 95% CI 0.85 to 1.18) (n = 5 RCTs) and based on observational comparative studies (RR = 1.10, 95% CI 1.01 to 1.21) (n = 8 studies). Caution is advised in the interpretation of these results given that the overall risk of bias in the RCTs is unclear due to limited available study details (see Results of critical appraisal of study methodology). Furthermore, although heterogeneity was low and not statistically significant, we note the presence of clinical heterogeneity across the studies.

Figure 10 in Appendix 4 reports pooled study estimates from a random-effects meta-analysis of LATP-any versus LATRUS for detection of CS prostate cancer. There is no statistically significant difference between LATP-any biopsy and LATRUS biopsy in detection of CS prostate cancer, based on a single RCT (RR = 1.14, 95% CI 0.65 to 2.01) and based on observational comparative studies (RR 1.20, 95% CI 0.98 to 1.47) (n = 4 studies).

Prostate cancer detection (local anaesthetic transperineal-any vs. general anaesthetic transperineal grid and stepping device, decision question 1)

Table 9 reports study cancer detection rates from the four studies which compared LATP-any biopsy versus GATP biopsy using grid and stepping device, and Figure 11 in Appendix 4 shows a meta-analysis forest plot containing three of the four studies (NB: the study publication by Walters et al. did not provide numerical cancer detection rates and was therefore not included in the meta-analysis). 44 There was some inconsistency between the studies in the direction of effects, with two studies marginally favouring LATP-any42,45 and another (smaller) study showing a large effect in favour of GATP. 43 Overall, there is no statistically significant difference between the two biopsy modalities in detection of prostate cancer, as estimated by a single RCT (RR = 1.05, 95% CI 0.76 to 1.44) and observational comparative evidence (RR = 0.76, 95% CI 0.34 to 1.72) (n = 2 studies).

Prostate cancer detection (network meta-analysis of local anaesthetic transperineal-any vs. local anaesthetic transrectal ultrasound vs. general anaesthetic transperineal grid and stepping device, decision question 1)

We used MetaInsight software23 to conduct a frequentist random-effects NMA of cancer detection rates for the biopsy modalities relevant to decision question 1 (see Figure 12, Appendix 4). The NMA provides an indirect comparison between LATP-any, LATRUS and GATP grid and stepping device to provide clinical effect estimates used in our economic analysis (see Model parameters). We restricted the NMA to RCTs because, in principle, randomised study designs have greater internal validity than observational studies (notwithstanding the uncertain risk of bias in RCTs we discussed earlier– see Results of critical appraisal of study methodology).

Consistent with the results of the pairwise meta-analyses above, the results of the NMA show RRs just below or just above RR = 1 for the respective biopsy modalities. Confidence intervals cross 1, indicating no statistically significant differences in cancer detection rates between the three biopsy modalities (see Figure 13, Appendix 4).

Our original economic base case included these NMA cancer detection rates in the economic model; however, at the request of NICE the base case was subsequently revised to exclude the Hara et al. 26 and Takenaka et al. 28 trials from the NMA. This was due to uncertainties raised by NICE Specialist Committee Members about the most appropriate anaesthesia classification for these studies (i.e. LATP or GATP) arising from the clinically atypical approach to anaesthesia administration taken (spinal injection in the transperineal trial arm and caudal block in the transrectal trial arm). The revised economic base case was accompanied by two economic scenario analyses: scenario 1 in which Hara et al. 26 and Takenaka et al. 28 were retained in the NMA as originally classified (i.e. LATP, as Figure 13, Appendix 4); and scenario 2 in which Hara et al. 26 and Takenaka et al. 28 were reclassified as GATP. The results for all three NMAs are given in Table 31.

Prostate cancer detection (local anaesthetic transperineal-freehand vs. local anaesthetic transrectal ultrasound, decision question 2)

Cancer detection rates, including CS cancer rates (where available), for six of the seven studies comparing LATP-freehand versus LATRUS are reported in Table 10 (NB: the remaining study36,37 did not report cancer detection as an outcome). The PrecisionPoint freehand device was evaluated in all six studies, and collectively the studies comprise a subset of LATP-any studies for decision question 1 presented earlier.

Figure 14 in Appendix 4 reports results of the random-effects meta-analysis of cancer detection rates for LATP-freehand versus LATRUS. A borderline non-statistically significant difference favouring LATP-freehand was estimated by the single relevant RCT (RR = 1.40, 95% CI 0.96 to 2.04). A statistically significant difference favouring LATP-freehand was estimated by the pooled observational comparative studies (RR = 1.21, 95% CI 1.08 to 1.34) (n = 4 studies).

To permit an incremental assessment of biopsy modality effects in our economic model we considered splitting the ‘LAPT-any’ study category into its constituent biopsy subtypes, that is LATP-freehand, LATP grid and stepping device and LATP coaxial needle (double freehand). However, details of biopsy procedures were limited in some study publications and it was unclear whether studies of LATP grid and stepping device or LATP coaxial needle (double freehand) could be reliably classified as such. Hence, as a pragmatic adjustment to allow an assessment of incremental cost-effectiveness, we combined these two biopsy modalities into a general category we refer to as ‘LATP-other’. It should be acknowledged, however, that a potential limitation is the underlying assumption that LATP using a grid and stepping device and LATP with a coaxial needle are necessarily equivalent in effects.

When RCT and observational evidence are pooled in our exploratory meta-analysis there is a statistically significant effect in favour of LATP-freehand compared to LATRUS (RR 1.22, 95% CI 1.10 to 1.35) (see Figure 13, Appendix 4). In contrast, there is no statistically significant difference between LATP-other and LATRUS for RCT evidence or observational comparative evidence or the two combined (see Figure 15, Appendix 4).

In the random-effects meta-analysis of LATP-freehand compared to LATRUS for CS prostate cancer detection, RRs non-significantly favoured LATP-freehand, as based on a single RCT (RR = 1.14, 95% CI 0.65 to 2.01) (see Figure 16, Appendix 4). There was a borderline statistically significant difference favouring LATP-freehand based on the observational comparative studies (RR = 1.31, 95% CI 1.00 to 1.72) (n = 3 studies). When observational and RCT studies are pooled in our exploratory analysis, a statistically significant difference is estimated (see Figure 16, Appendix 4).

Prostate cancer detection (local anaesthetic transperineal-freehand vs. general anaesthetic transperineal grid and stepping device decision question 2)

A single study compared cancer detection rates between LATP-freehand (PrecisionPoint) versus GATP grid and stepping device. 45 The study is a retrospective review of people who underwent transperineal prostate biopsy under local anaesthetic or under general anaesthetic, performed by a single surgeon. There was a small difference of seven percentage points in cancer detection rates, favouring PrecisionPoint [cancers detected: 65/72 (90.0%) PrecisionPoint versus 59/71 (83.0%) GATP]. [NB; this is one of the studies included in the comparison of LATP-any vs. GATP grid and stepping device presented earlier: see Prostate cancer detection (LATP-any vs. GATP grid and stepping device, decision question 1)].

Prostate cancer detection (network meta-analysis of local anaesthetic transperineal-freehand vs. local anaesthetic transperineal-other vs. local anaesthetic transrectal ultrasound vs. general anaesthetic transperineal grid and stepping device, decision question 2)

We used MetaInsight software (Owen et al.)23 to conduct a frequentist random-effects NMA of cancer detection rates from RCTs for decision question 2 (see Figure 17, Appendix 4). This provided an indirect comparison between LATP-freehand versus LATP-other versus LATRUS versus GATP grid and stepping device, to inform an incremental assessment of cost-effectiveness in our economic analysis (see Model parameters).

Prostate cancer detection risk classification

Table 11 compares risk classification scores for people with detected prostate cancer biopsy for LATP-any versus LATRUS. The risk of the prostate cancer progressing aggressively was commonly assessed using Gleason scores (higher scores indicate greater progression risk), though other classification systems appear to have been used. 34 Not all studies provided risk classification for the comparator biopsy arm, but where comparative data were given Gleason scores were similar. Two of the studies34,41 are also relevant to the comparison of LATP-freehand versus LATRUS (decision question 2). 34,35,41

A single (retrospective observational) study reported cancer risk classification for the comparison of LATP-any versus GATP grid and stepping device. 45 The study used the International Society of Urological Pathology (ISUP) grade group classification as ‘low risk’ to ‘Intermediate Favourable risk’. The LATP biopsy was done using the PrecisionPoint freehand device, thus this study is also relevant to ‘LATP-freehand versus GATP grid and stepping device (decision question 2)’. A higher percentage of participants were classified as ISUP > 2 by the LATP biopsy than GATP, but this was not statistically significant [n = 35/65 (53.8%) vs. n = 28/59 (47.5%), respectively, p = 0.48].

Diagnostic accuracy of prostate biopsy

None of the included studies fully reported the diagnostic or prognostic accuracy of LATP biopsy. Rather, as mentioned earlier, studies tended to report cancer detection rates without necessarily verifying the accuracy of cancer detected against a reference standard in terms of measures such as sensitivity and specificity.

One study reported the proportion of all cancers detected under LATP and under GATP (clinical sensitivity), but did not provide information on proportion of cancers not detected (clinical specificity). 45 A reference standard was not reported either. This study is currently available only as a conference abstract, hence limited information.

Another study reported the pathological accordance of Gleason scores based on biopsy with histological analysis of prostatectomy specimens (i.e. a reference standard). 43 This resulted in a small proportion of participants having their Gleason scores upgraded and upstaged.

Hospitalisation events after biopsy

Hospitalisation following prostate biopsy was reported by a total of 10 studies, for 4 of the 5 biopsy comparisons relevant to the decision problem (see Tables 12–14). Studies tended to report the number of participants admitted to hospital at various time points after the biopsy (e.g. up to 30 days post biopsy), while others reported hospitalisation in response to serious complications such as fever and pneumonia. Less commonly reported was the duration of hospital stay. Overall, rates of hospitalisation were numerically higher for comparator biopsy approaches compared to LATP across the four biopsy comparisons. However, hospitalisation rates were very low in general, and it is therefore difficult to make definitive conclusions on the currently available evidence.

The cost of hospital stays can be influential in the assessment of cost-effectiveness of health care. We discuss the hospitalisation estimates which inform our economic analysis of prostate biopsy in Biopsy-related complications.

Overall biopsy-related complications

Six studies reported overall rates of complications following prostate biopsy. Some, but not all, of the studies reported overall rates in addition to rates of the constituent complications. We report here only studies which presented an overall complication rate; we did not sum rates of specific named complications to create an overall total complication rate for each study. All six studies were comparisons of LATP-any biopsy versus LATRUS biopsy and are relevant to decision question 1 (Table 15). Two of the six studies compared freehand transperineal devices versus LATRUS and therefore are also relevant to decision question 2. 30,31,34,35

Specific biopsy-related complications

Rates of urinary retention

Post-biopsy urinary retention is reported by nine studies in total across three biopsy comparisons (see Tables 22–24). Some studies reported retention data for the LATP biopsy but not the comparator. Where comparative evidence was available, retention rates were similar between biopsy modalities, though it is difficult to make definitive conclusions based on small event rates.

No studies reported post-biopsy urinary retention for the comparison of LATP-freehand versus GATP/LATP using a grid and stepping device (decision question 2).

Rates of erectile dysfunction

Only two studies in this systematic review reported assessing post-biopsy erectile dysfunction. 27,33 Both used the International Index of Erectile Function (IIEF-5) instrument, in which lower scores indicate greater severity of erectile dysfunction. The observational study by Hung et al. reports that mean IIEF-5 change post biopsy was 2.74 in LATRUS and 6.03 in LATP, and was statistically significant (p = 0.023). 33

The RCT by Lam et al. reports a reduction in the IIEF-5 score that was ‘more significant in LATP arm’ p < 0.05. 27 No further detail is given to quantify this statement. Details of these two studies are publicly available only as a conference abstract at the time of writing. The EAG has been told, via personal communication with the lead investigator,27 that a manuscript is being prepared for submission to a journal.

Survival

None of the included studies reported survival outcomes for participants receiving biopsy.

Progression-free survival

None of the included studies reported progression-free survival for participants treated for prostate cancer detected on biopsy.

Adverse events from treatment

None of the included studies reported adverse events in participants treated for prostate cancer detected on biopsy.

Patient-reported tolerability

A total of 12 studies reported data on the degree of pain and discomfort during prostate biopsy as rated by patients (see Tables 25 and 26). Tolerability was measured in a variety of ways across the studies, but often data are only presented for the LATP biopsy group, thus limiting comparisons to be drawn between types of biopsy.

Methods for review of economic studies

The database searches for cost-effectiveness were carried out on 17 June 2021 and updated on 2 November 2021. The search strategies were based on an early version of the clinical-effectiveness searches with the addition of the Canadian Agency for Drugs and Technologies in Health (CADTH) filter for Economic Evaluations/Cost/Economic Models applied to the MEDLINE and EMBASE strategies and amended versions of the filter applied to the Cochrane Library and Web of Science strategies. 60 The INAHTA, DARE and NHS EED strategies were the same as for the clinical-effectiveness searches. In addition, the EconLit database was searched. An English-language limit was applied. The full search strategies are shown in Appendix 1 (Table 47).

The relevant population, interventions and comparators are the same as for the systematic review of test yield and clinical effectiveness (see Inclusion and exclusion criteria) but differed in terms of the relevant study design and outcomes. Studies were included if they were full economic evaluations, assessing both costs and consequences, for the specified diagnostic strategies. Outcomes included are those consistent with full economic evaluations, including measures of resource use and costs, and health outcomes: life-years (LYs) or quality-adjusted life-years (QALYs) gained. Each step of the review was completed by two health economists and any disagreements were resolved by discussion. Studies that reported resource use, costs or HRQoL in the area of prostate cancer were excluded if they did not meet the inclusion criteria above, but considered separately as possible sources of evidence to inform model structure and inputs.

The EAG planned to extract data related to the study design, methods, parameter sources, relevant model inputs and results of the included cost-effectiveness studies. The credibility of the included cost-effectiveness studies and their relevance to current UK practice were assessed using a pre-defined checklist, shown in Appendix 6, Table 64. This checklist was based on the International Society for Pharmacoeconomics and Outcomes Research (ISPOR)61 and Philips et al.'62 checklists.

Results of the review of economic studies

Starting with 724 potentially relevant references identified in the original (704) and updated (20) searches, 10 studies were retrieved for full-text screening (see flowchart in Appendix 6, Figure 19). After inspection, nine references were excluded (see Appendix 6, Table 62 for the reasons for exclusion).

Overview of other published economic studies of interest

Other studies retrieved by the systematic review were considered as possible sources of evidence to inform our model structure and inputs (see Appendix 6, Table 65). Most of these studies are evaluations of the use of mpMRI to inform TRUS biopsies versus TRUS alone in people with suspected prostate cancer, a prior negative or inconclusive biopsy or undergoing active surveillance. The remaining studies assessed screening or other diagnostic tests and assays (vs. TRUS or a PSA test) in men with suspected prostate cancer. Eight out of 13 studies used a decision tree plus a Markov model, while 2 used a decision tree only and another two used a Markov model only. One of the studies used a microsimulation model. Most studies applied a lifetime horizon and a 1-year Markov cycle length. All the studies reported costs and utilities and estimated the cost per QALY gained. Two economic studies were very influential in the development of our model and are discussed below.

Methods for review of health-related quality of life

We undertook searches to identify data on HRQoL for patients undergoing screening and diagnosis of prostate cancer, and for patients with diagnosed prostate cancer. The aim of these searches was to identify utility values that were suitable for use in the economic model.

A sequential approach was used to identify HRQoL studies:

1. Systematic searches of bibliographic databases were conducted for HRQoL data in people with suspected prostate cancer (searches ‘HRQoL 1’).

2. Additional systematic searches of bibliographic databases were conducted for HRQoL data in people with both suspected as well as diagnosed prostate cancer (searches ‘HRQoL 2’), to find additional utility values suitable for the economic model not identified in the ‘HRQoL 1’ searches.

The first set of searches (HRQoL 1) used the clinical-effectiveness search strategies with the addition of the CADTH search filter for Health Utilities/Quality of Life applied to the MEDLINE and EMBASE strategies and amended versions of the filter applied to the Cochrane Library and Web of Science strategies. The second set of database searches (HRQoL 2) were subsequently run with the biopsy terms removed to retrieve studies that would cover the whole disease pathway in addition to the diagnostic process. In order to save time, HRQoL 2 included terms specific to the European Quality of Life Working Group Health Status Measure 5 Dimensions (EQ-5D) utility measure (the CADTH search filter was not used), to reflect the NICE preferred method for utility assessment,68 with the option to expand the search to other utility measures if needed. The searches were carried out in MEDLINE, EMBASE, Web of Science and the Cochrane Library, and they were limited to the most recent 10 years. The strategies for HRQoL 1 and HRQoL 2 are shown in Appendix 1 (see Tables 48 and 49).

The inclusion and exclusion criteria for eligibility screening are given in Appendix 7, Table 66. The same eligibility criteria were used for screening both titles and abstracts and full-text records. Only primary research studies were included. The relevant population was people who had undergone screening or diagnostic tests for prostate cancer, or with diagnosed prostate cancer. We planned to extract data related to the study design, country and sample size, HRQoL instruments used, and health states assessed.

Results of the review of health-related quality-of-life studies

The systematic searches ‘HRQoL 1’ identified 244 potentially relevant studies (see Appendix 7, Figure 20). Of the 244 references, 34 were retrieved for full-text screening and 9 studies were included. 69–77 The excluded references and reasons for exclusion are shown in Appendix 7, Table 67. Study characteristics and results from the included studies are summarised in Appendix 7, Tables 68 and 69. However, these studies did not provide suitable utility values for our economic model.

The systematic searches ‘HRQoL 2’ identified 369 potentially relevant studies, of which 21 were retrieved for full-text screening, and 6 studies78–83 were included (see Appendix 7, Figure 21). The excluded references and reasons for exclusion are shown in Appendix 7 (Table 70).

The main characteristics and results of the six studies included from ‘HRQoL 2’ are presented in Tables 27 and 28 (further detail in Appendix 7, Table 71). Three studies were conducted in the UK and used the EQ-5D-5L, and three were conducted in Finland, from which two used the EQ-5D-3L version with a UK tariff and the other did not specify the version used. Overall, the studies reported EQ-5D scores associated with no cancer, early/localised prostate cancer and late/metastatic prostate cancer. All the studies, except one, have a sample size > 300. These papers are discussed in relation to their applicability to the EAG economic model in Utilities.

The modelled cohort

The base case population entering the model is a cohort referred for a first prostate biopsy for suspected localised prostate cancer after mpMRI with Likert score of 3 or more. We also conduct analysis for three other subgroups: mpMRI Likert score of 1 or 2 at first biopsy; mpMRI Likert score of 3 or more after a previous negative biopsy; and mpMRI Likert score of 1 or 2 after a previous negative biopsy. For our base case, we assume that there are no people with metastatic prostate cancer in the cohort because it is likely that people with overt MD and those for whom active treatment for diagnosed disease would not be appropriate would have been screened out of the cohort prior to biopsy. We tested the impact of including a proportion of people with pre-existing MD in scenario analysis (see Appendix 9, Table 88).

The diagnostic pathway: decision tree

The structure of the decision tree is described in detail in Decision tree structure. The design and parameter sources are largely based on the PROMIS economic analysis reported by Faria and colleagues, and the version of this analysis adapted by Wilson et al. to estimate cost-effectiveness for LATP (see Results of the review of economic studies and Overview of other published economic studies of interest above for discussion of these studies). 63–65

The cohort entering the decision tree is first stratified by baseline prevalence of LR, IR and HR localised disease, and MD (if included). The tree then models the diagnostic pathway with the alternative biopsy methods specified in the scope, and estimates resulting complication and cancer detection rates, costs and QALYs. The tree includes a second biopsy for a proportion of patients with a negative first biopsy, with the assumption that the second biopsy would be conducted with the LATRUS method. This is a simplification. In practice, methods for repeat biopsies are likely to vary, but evidence for the diagnostic yield of other biopsy methods after a previous negative first biopsy is sparse. The proportion undergoing repeat biopsy can be changed.

Inputs to the decision tree are:

• Baseline prevalence stratified by level of risk, estimated from data reported by Faria et al. (see Baseline prevalence). 64,65,94

• Probabilities of detecting CS and clinically non-significant (CNS) prostate cancer (see Cancer detection rates). For LATRUS, these probabilities are also estimated from data reported by Faria et al. 64,65,94 These baseline probabilities are adjusted for other biopsy methods using RR estimates from the EAG systematic reviews and meta-analyses (see Intermediate outcomes).

• The probability of a repeat biopsy if the first biopsy is negative is estimated from a paper by Jimenez et al., identified from our clinical review. 85 Assumptions about how this probability differs according to the first biopsy method and result were tested in scenario analysis (see discussion in Probability of a repeat biopsy).

• Probabilities of biopsy-related complications were estimated from various sources. 66,96–100 Relevant papers were identified from our clinical review and the review of economic evaluations (see Clinical outcomes, Results of the review of economic studies and Overview of other published economic studies of interest), with alternative sources tested in scenario analysis (see discussion in Biopsy-related complications).

• Costs of the biopsy procedures and treatment for complications, see Costs of the biopsy procedure and Resource use and costs for management of suspected prostate cancer. We developed detailed cost estimates for different LATP approaches in decision question 2.

• The impact of biopsy-related complications on patient HRQoL and survival (QALY loss) is based on assumptions as in the analysis by Wilson et al. (see Utilities). 63

Long-term consequences: the Markov model

We considered two designs for the Markov model based on existing studies:

1. a model with three health states (progression-free, MD and death), stratified by initial level of cancer risk and treatment, developed for the PROMIS economic evaluation by Faria et al.; and

2. a model developed by the NICE Guideline Updates Team for the 2019 update of the NICE prostate cancer guideline (NG131) evaluation of follow-up strategies for people with a negative mpMRI or biopsy result. 64,65,67

The NG131 model structure includes some features that make it more appropriate for the current decision problem. In particular, it explicitly models subsequent diagnosis for people with FN results after the biopsy pathway, based on estimated rates of symptomatic presentation and routine follow-up in primary care. This enables quantification of the monetary and QALY costs of a biopsy failing to diagnose CS disease and the resulting delay in treatment. The NG131 model also includes costs for diagnosis and follow-up and a wider range of treatments that reflect NICE guidance. We therefore decided to adapt the NG131 Markov model structure for our analysis.

The structure and input parameters of the NG131 model are described in the health economic model report available on the NICE website. 67 We also had access to a copy of the model. Input parameters and assumptions in our model were aligned with those in the NG131 model, except where more recent or relevant sources were identified. We included parameters to specify a schedule of primary care follow-up for people with one or more negative biopsy result. For our base-case analysis, we assumed a PSA test at 6 months, then annual tests for a maximum of 10 years for everyone with a FN biopsy result. Modelled treatment for diagnosed prostate cancer reflected NICE guidance at the time of the 2019 guideline update, with additional treatments for MD based on more recent technology appraisals. 86–88

See Markov model structure for further description of the NG131 model structure and explanation of how we adapted it for the current decision problem.

Parameters for the Markov model include:

• Transition probabilities between the 11 health states. We used the same probabilities as in the NG131 model (Table HE07 in the online model report). 67 See Long-term transition probabilities for details and explanation of how these probabilities were derived.

• Resource use and costs, including monitoring and follow-up, treatment of diagnosed prostate cancer, management of adverse events and end-of-life costs (see Resource use and costs for management of suspected prostate cancer for details).

• Health outcomes are estimated in the form of QALYs, incorporating modelled survival and the impact of symptoms and adverse effects on utility. 91,92,109 See Utilities.

Framework for economic analysis

Analysis followed the NICE reference case at the time of analysis, as specified in section 15 of the Diagnostics Assessment Programme (DAP) manual. 68

• The model uses a ‘lifetime’ time horizon (up to a maximum age of 100 years) to reflect the life-threatening consequences of misdiagnoses or serious biopsy-related complications. The Markov model uses a 3-month model cycle.

• Health outcomes are estimated as QALYs, with utilities estimated from EQ-5D-3L data with NICE-recommended UK general population values.

• Costs are estimated from an NHS and personal social services perspective. Biopsy costs are estimated with a micro-costing approach, informed by company submissions and expert judgement. Unit costs are taken from standard national and NHS sources. 89,90 The base case uses long-term average cost estimates for the interventions and comparators, with annuitised costs for capital equipment.

• Standard rates of discounting for time preference over costs and QALYs are applied, as recommended by NICE (currently 3.5% per year for costs and QALYs).

Population and subgroups

The model was designed to estimate costs and health outcomes for the population specified in the NICE scope: people with suspected prostate cancer referred for prostate biopsy. We aimed to reflect characteristics of this population in routine NHS practice, including age and probability of prostate cancer prior to biopsy.

The National Prostate Cancer Audit (NPCA) reported that 54% of people newly diagnosed with prostate cancer in England and Wales between April 2018 and March 2019 were aged 70 years or over (mean age at diagnosis was not reported) (NPCA 2020 Table 3). 92 However, one would expect the mean age at biopsy to be lower than the mean age at diagnosis. The mean age at referral for a first prostate biopsy in the PROMIS study was 63.4 years, but the mean age for those diagnosed with IR and HR cancers was 64.9 and 66.8 years, respectively. 64 For the base case, we assumed a mean age of 66 years at referral for biopsy, as this matches the assumption in the NG131 update analysis, as well as feedback from a specialist committee member. 67 We tested the effect of baseline age in scenario analysis (see Appendix 9, Table 88).

For the purposes of the economic evaluation, we assumed that the cohort had already had mpMRI as an investigation for suspected clinically localised prostate cancer prior to referral for biopsy, with results reported on a five-point Likert scale. This aligns with the NICE recommendation from the 2019 update of NG131 (recommendation 1.2.2). 8 Use of the Likert scale is also consistent with evidence of the diagnostic yield of mpMRI available from the PROMIS study, which we use to estimate the baseline prevalence of prostate cancer conditional on mpMRI results (see Baseline prevalence). We acknowledge that this does not necessarily align with clinical practice, as some centres use PI-RADS instead of Likert to report mpMRI results. There is also uncertainty over the generalisability of evidence on the comparative diagnostic yield of biopsy methods, as some studies did not report prior mpMRI use, and those that did report results in terms of PI-RADS rather than Likert scores (see Characteristics of studies comparing LATP prostate biopsy by any method versus LATRUS prostate biopsy (decision question 1), Characteristics of studies comparing LATP prostate biopsy by any method versus GATP prostate biopsy using a grid and stepping device (decision question 1), Characteristics of studies comparing LATP prostate biopsy using a freehand device versus LATRUS prostate biopsy (decision question 2), Characteristics of studies comparing LATP prostate biopsy using a freehand device versus GATP prostate biopsy by grid and stepping device (decision question 2) and Characteristics of single-arm studies evaluating LATP prostate biopsy using a freehand device where no comparative evidence was identified).

Faria et al. 65 evaluated ‘true disease’ status for the PROMIS population based on a combination of a TPM biopsy and TRUS biopsy (whichever was the most severe). They classified LR, IR and HR localised prostate cancer according to two sets of definitions. For the economic model, we used the following:

• LR: Gleason ≤ 6, PSA ≤ 10 ng/ml and clinical stage T1 to T2a

• IR: Gleason 7, PSA 10–20 ng/ml and clinical stage T2b

• HR: Gleason 8–10, PSA > 20 ng/ml and clinical stage T2c or higher.

Intermediate- and high-risk localised disease are grouped together as CS disease. LR disease is classed as CNS.

We assume that the referred cohort does not include people with MD. NICE guidance is that people who are not going to be able to have radical treatment should not be routinely offered mpMRI (NG131 recommendation 1.2.1), and that those for whom clinical suspicion of prostate cancer is high because of high PSA value and evidence of bone metastases should not be routinely offered prostate biopsy for histological confirmation (NG131 recommendation 1.2.8). The PROMIS, which provides baseline estimates of prevalence for the model, excluded people with MD; 5 out of 740 men registered for the study were withdrawn due to having stage T4 or nodal disease (Brown et al., table 6). 64

Patient subgroups

In our base-case analysis, we focus on people referred for a first biopsy with a prior mpMRI Likert score of 3 or more (NG131 recommendation 1.2.3). NG131 recommends considering omission of a prostate biopsy for people with a mpMRI Likert score of 1 or 2, but only as a shared decision after discussion of the risks and benefits with the person concerned (NG131 recommendation 1.2.4). The NICE scope for the current assessment reports expert opinion that around 40% of people with Likert score of 1 or 2 are discharged based on the results of the mpMRI scan. This group are less likely to have CS prostate cancer than those with a mpMRI score of 3 or more. Similarly, the risk of prostate cancer, and hence cost-effectiveness, is likely to differ for people who have never had a prostate biopsy, and for those who have had a previous negative prostate biopsy and are referred back.

We assess cost-effectiveness separately for the following subgroups:

1. people referred for a first biopsy with a Likert score of 3 or more (base case)

2. people referred for a first biopsy with a Likert score of 1 or 2

3. people referred after a previous negative biopsy with a Likert score of 3 or more

4. people referred after a previous negative biopsy with a Likert score of 1 or 2.

We do not present subgroup analysis by location of lesions or enlarged prostate, due to a lack of evidence to differentiate prognosis or diagnostic yield for these groups.

The model uses prevalence of LR, IR and HR localised prostate cancer in subgroups A to D estimated from data on true disease status in the PROMIS cohort and diagnostic yield characteristics of mpMRI and TRUS biopsy reported by Faria et al. See Baseline prevalence for details of the prevalence calculations.

Biopsy methods and devices

The model was designed to evaluate the decision questions defined in the NICE scope. Following the naming conventions for interventions and comparators used in the pairwise and network meta-analyses in Intermediate outcomes, we conducted the following comparisons.

Decision question 1:

• LATP prostate biopsy with a freehand device, grid and stepping device or coaxial needle (LATP-any)

• LATRUS biopsy (LATRUS)

• GATP using a grid and stepping device (GATP).

Decision question 2:

• LATP prostate biopsy with a freehand device (LATP-freehand)

• LATP prostate biopsy without a freehand device (LATP-other), including LATP conducted with a grid and stepping device or coaxial needle

• LATRUS biopsy (LATRUS)

• GATP with a grid and stepping device (GATP).

Decision-tree structure

A simplified overview of the decision tree is shown in Figure 4. This tree is replicated for each biopsy method under comparison.

FIGURE 4.

Overview of decision tree. CNS, clinically non-significant cancer; HR, true high-risk; HR Dx, high-risk correctly diagnosed; HR FN, misdiagnosed intermediate-risk; IR, true intermediate-risk; IR Dx, intermediate-risk correctly diagnosed; IR FN, misdiagnosed intermediate-risk; LR, true low-risk; LR Dx, low-risk correctly diagnosed; LR FN, misdiagnosed low-risk; MS, true metastatic disease; MS Dx, metastatic correctly diagnosed; MS FN, misdiagnosed metastatic; NC Dx, no cancer correctly diagnosed.

The model starts with a cohort of interest, one of four subgroups A–D defined by mpMRI Likert score and history of previous biopsy. The cohort is first stratified by true prostate cancer status (no cancer, LR, IR or HR localised disease or MD). The decision tree then estimates diagnostic outcomes (the proportions of correct and FN biopsy results) and complications from the biopsy process. The end points of the decision tree, correct diagnoses (Dx) or FNs, represent the initial health states in the Markov model.

Biopsy-related complications are categorised as:

• no adverse event: no or minor events for which the patient does not seek treatment

• mild adverse events: mild/moderate events treated outside hospital

• admission: admission within 28 days of the biopsy

• mortality within 28 days of the biopsy.

Costs are estimated for the biopsy process, including costs of the initial and repeat biopsies, and costs for management of any biopsy-related complications. QALYs accumulated within the biopsy process are also estimated, based on initial utility values assigned to the cohort and allowing for any QALY loss attributable to complications, including disutility for transient adverse events and lifetime QALY loss for rare biopsy-related deaths.

The following sections describe the structure of the subtrees for people without prostate cancer (NC) and for those with CNS (LR) or CS (IR, HR) localised prostate cancer. The model also includes a subtree for MD, but this is not used in the current analysis.

No prostate cancer

See Figure 5 for an illustration of the biopsy process for people who do not have prostate cancer. We assume that all biopsy methods are perfectly specific: so there cannot be false positive results for people who truly do not have prostate cancer. However, it is possible that a patient may be referred for a repeat biopsy if there is a high level of clinical suspicion. Biopsy-related complications may occur, classified as above (mild, admission or mortality). End points for people without prostate cancer are correct diagnosis (NC Dx) or death from biopsy-related complications.

FIGURE 5.

Illustration of decision tree for people without prostate cancer. AE, adverse events; Bx, biopsy; NC Dx, no cancer correctly diagnosed.

Clinically non-significant prostate cancer

Figure 6 illustrates the biopsy process for people with CNS disease (LR prostate cancer). For this population, the biopsy may give a correct diagnosis, a false positive result of CS disease or a FN result of no cancer. In practice, there were no cases of people with LR cancer receiving a CS biopsy result in the PROMIS study (see Table 30). 64 Hence, the probability of this event in our model is zero.

FIGURE 6.

Illustration of decision tree for people with LR prostate cancer. AE, adverse event; Bx, biopsy; LR, true low-risk; LR Dx, low-risk correctly diagnosed (classified as CNS); LR FN, low-risk false negative (classified as no cancer); LR FP, low-risk false positive (classified as CS).

If the biopsy is negative (CNS or no cancer), a repeat biopsy may be performed. We assumed that the probability of a repeat biopsy is higher if the result of the first biopsy is CNS or if the prior mpMRI Likert score was 3 or more, than if the first biopsy result is ‘no cancer’ and the Likert score is ˂ 3. A second biopsy may give a CS, CNS or no cancer result, although the estimated probability of a CS result for a second TRUS biopsy with LR cancer is zero (as in the PROMIS model, based on the systematic review and meta-analysis by Schoots et al.). 65,93

Complications may occur after the first and/or second biopsy, classified as above (none, mild, admission or mortality). End points for people with LR disease are correct diagnosis (LR Dx), false positive (LR FP), false negative (LR FN) or death.

Clinically significant prostate cancer

The structure of the decision tree is the same for people with IR prostate cancer (illustrated in Figure 7) as for those with HR disease (figure not shown). We assume that the incidence of complications does not differ by cancer risk group. End points for intermediate and high risk are correct diagnosis (IR Dx; HR Dx), false negative (IR FN; HR FN) or death.

FIGURE 7.

Illustration of decision tree for people with IR prostate cancer. AE, adverse event; Bx, biopsy; IR, true intermediate-risk; IR Dx, intermediate-risk correctly diagnosed (classified as CS); IR FN, intermediate-risk FN (classified as CNS or no cancer).

Markov model structure

As discussed above (see Long-term consequences: the Markov model), we considered two designs for the Markov model to estimate long-term costs and QALYs from the diagnostic outcomes from our decision tree: the model developed for the economic evaluation of the PROMIS study by Faria et al., later adapted by Wilson et al.; and the model developed for the 2019 update of the NICE prostate cancer guideline (NG131). 63–65,67 We chose to use a replicated version of the NG131 Markov model as this gives a more flexible structure to model the costs and consequences of missed diagnoses, allowing for possible future diagnosis.

The structure of this model is illustrated in Figure 8. It includes health states based on prostate cancer and diagnostic status: no cancer (NC) and diagnosed and undiagnosed LR, IR and HR localised disease and MD. The initial distribution of the cohort across the Markov states is determined by the output from the biopsy pathway modelled in the decision tree: with some ‘overdiagnosis’ of LR disease and some missed diagnoses of CS disease. Rates of transition from undiagnosed to diagnosed health states can be set to reflect primary care monitoring and symptomatic presentation. The model includes simplified assumptions about sequential progression of prostate cancer: from incident LR disease through IR and HR localised disease to MD. Death from prostate cancer is assumed to only occur with MD, although death from other causes occurs from all states.

FIGURE 8.

Schematic of Markov model structure. PC, prostate cancer.

Baseline prevalence

Estimates of the true prevalence of prostate cancer for each of the subgroups, A–D, are shown in Table 29. These provide the starting proportions of the cohort allocated to LR, IR and HR disease in the decision-tree model. They were derived from the following PROMIS results reported by Faria et al.:65

• True cancer status in the PROMIS cohort defined by the most severe of the template mapping biopsy and TRUS biopsy results (Faria et al. supplement table 5). 65

• Diagnostic yield of mpMRI: probability of mpMRI result of no cancer, CNS or CS disease, given true cancer status (Faria et al. table 3, CNS definition 2, Likert cut-off ≥ 3). 65

• Diagnostic yield of first TRUS biopsy: probability of finding of no cancer, CNS or CS disease, given true cancer status (Faria et al. table 2, Type 4, CNS definition 2). 65

We combined these results using Bayes formula. For example, the probability that a member of subgroup A does not have cancer is calculated from the probability that someone with no cancer had Likert ≥ 3, the proportion of the cohort with no cancer and the overall probability of Likert ≥ 3: p(NC | Likert ≥ 3) = (p(Likert ≥ 3 | NC) * p(NC))/p(Likert ≥ 3).

We note the zero probability of true HR localised prostate cancer for people with a Likert 1 or 2 result from mpMRI (Faria et al., supplementary table 7). We understand that such cases may occur in practice, although this may reflect inaccurate mpMRI scoring.

Cancer detection rates

Probability of a repeat biopsy

The probability of patients having a second biopsy after a negative first biopsy is based on a prospective cohort study reported by Jimenez et al.. 85 They assessed whether an initial GATP biopsy (systematic TP with 30 cores taken in theatre under general or spinal anaesthetic) translated into a lower risk of re-biopsy compared with LATRUS (systematic transrectal biopsy with 12–18 cores taken in office under local anaesthetic). Repeat biopsy was indicated based on a protocol defined by the authors (see table 1 in Jimenez et al.). 85 The number of patients having GATP in the cohort was much smaller than those having LATRUS, and patients with larger prostates were preferably selected for GATP. During the study period, 15.5% (95/615) and 5.4% (3/56) of patients had repeat biopsies after LATRUS and GATP respectively (p = 0.06). This difference was statistically significant in a multivariate analysis with adjustment for PSA density (p = 0.03). However, there are uncertainties over the generalisability of this result due to the lower sample size for GATP and differences in prostate volume and the numbers of core samples. We applied the LATRUS re-biopsy rate (15.5%) in our base case for all biopsy methods and varied this in scenario analyses (see Appendix 9, Table 87).

Biopsy-related complications

Comparative rates of complications from our systematic review are reported in Clinical outcomes. The included studies reported a variety of adverse outcomes, but the studies were too heterogeneous for meta-analysis and the individual studies lacked power to reliably estimate adverse-event rates. We therefore used other observational sources identified from our clinical and economic searches to estimate complication rates for the model. See Table 32 for a summary of sources used in our base case analysis.

For admission and mortality, we used results from an analysis of NHS Hospital Episode Statistics by Tamhankar et al.. 94 They included all patients with a code of M702 (transperineal needle biopsy of prostate) or M703 (transrectal needle biopsy of prostate) between April 2008 and March 2019, and identified those who were readmitted or attended accident and emergency within 28 days after the biopsy. These data do not distinguish between transperineal biopsies conducted under local or general anaesthetic. We used results from the two most recent years (April 2017–March 2019), following advice that these are more reflective of current practice. In the 2017–9 cohort, non-elective admissions were lower after TP than after transrectal biopsy, but the difference was not statistically significant (3.54% and 3.74% respectively, p = 0.11). Infections were the main cause of non-elective admissions after transrectal biopsy, while urinary retention was the main cause after TP. Mortality within 28 days of the procedure was rare and similar after transperineal and transrectal biopsy (0.05% and 0.07%, respectively).

The decision-tree model also included biopsy-related complications treated outside hospital (‘mild’ adverse events) (see Decision tree structure). For this outcome, we used complication rates from different sources for transrectal and transperineal biopsies, as we could not identify a suitable source that included both. Rosario et al. reported healthcare contacts within 35 days of a TRUS biopsy for a prospective cohort of 1147 patients nested within the ProtecT trial. 95 Pepe and Aragona reported complications within 15–20 days of a TP in a single-centre study in Italy. 96 This cohort included 3000 patients who underwent biopsy under general or local anaesthetic, with 12, 18 or 24 or more biopsy cores.

In the base case, we assumed the same rates of complications for LATP and GATP. We have received conflicting expert advice over the relative incidence of complications for transperineal biopsies under local or general anaesthetic, although Pepe and Aragona reported that these rates were ‘superimposable’. 96

Scenario analyses for other estimates of biopsy-related hospital admission rates are reported in Scenario analysis: probability of biopsy complications. In addition to scenarios with admission rates from the Rosario and Pepe and Aragona studies (as reported in Table 32), we report scenarios based on a study by Berry et al.. 97

Berry et al. used data from the NPCA linked to Hospital Episode Statistics to compare readmission rates within 30 days of a transrectal or TP (type of anaesthesia not reported) conducted prior to a new diagnosis of prostate cancer. People who underwent a TP were less likely to be readmitted to hospital because of sepsis, but more likely to be readmitted because of urinary retention than patients who underwent a transrectal biopsy. The analysis also showed that an overnight stay was significantly more likely immediately after a TP than after a transrectal biopsy (12.25% and 2.36%, respectively, p < 0.001). However, NICE specialist committee members advised that this difference is not reflective of current practice, as the Berry et al. analysis used data from a period prior to March 2017 when TP was conducted under general anaesthetic.

Long-term transition probabilities

Transition probabilities for the Markov model were based on values used in the NICE 2019 guideline update NG131. 8 The natural history parameters used to calculate transition probabilities are reported in Table HE07 of the health economic model report available on the NICE website. 67

The base-case transition probabilities (per 3-month model cycle) are shown in Table 33. The matrix differs for model cycles in which primary care follow-up (PSA testing and LATRUS biopsy if indicated) is expected for people with a negative diagnosis, because the probability of diagnosis for FN cases is higher than in the other model cycles, when diagnosis is only related to symptomatic presentation.

Costs of the biopsy procedure

We estimated the cost of each biopsy method using a micro-costing approach, including the following components:

• cost of device (where applicable)

• cost of general consumables (needles, antibiotics, anaesthesia, ultrasound, lithotomy bed, etc.)

• staff time for training

• staff time to perform biopsy (urologists, nurses, anaesthetists)

• cost of the place of biopsy (outpatient room, theatre session)

• cost of reprocessing for reusable devices

• cost of histopathological analysis

• urologist consultation to discuss biopsy results and disease management.

Estimates were based on information provided to NICE by the companies (including the YHEC study),84 from clinical experts, and from the study by Wilson et al.. 63 Where information was not available, we made assumptions. More details on the assumptions used in the estimation of biopsy costs are shown in Appendix 8. Costs of consumables are summarised in Appendix 8, Table 72.

We estimated a cost of £681 for LATRUS biopsy and £1251 for GATP (Table 34). The estimated cost varies between LATP methods and devices.

• For decision question 1, we used a base case cost of £776 for LATP-any, which is the mean of the named LATP devices (CamPROBE, PrecisionPoint, EZY-PA3, UA1232, Trinity Perine and SureFire Guide), LATP with a grid and stepping device and LATP with a coaxial needle (‘double freehand’).

• For decision question 2, we used a cost of £781 for ‘LATP freehand’: the mean cost for all named LATP devices (CamPROBE, PrecisionPoint, EZY-PA3, UA1232, Trinity Perine and SureFire Guide). For ‘LATP other’, we used the estimated cost for LATP with a grid and stepping device (£791).

See Scenario analysis: biopsy costs for scenario analysis with different estimates of biopsy costs.

Resource use and costs for management of suspected prostate cancer

In addition to the cost of biopsies, the model includes costs for subsequent follow-up and monitoring, diagnosis and the treatment of prostate cancer and adverse events. In this section, we summarise the key assumptions used for costing. Full details of resource use inputs to the base-case model are listed in Appendix 8, Table 73. Unit costs are listed in Appendix 8, Table 74.

Utilities

Health-related quality-of-life (utility) values for the decision model were derived from studies identified from the systematic review ‘HRQoL 2’ (see Tables 27 and 28) and from references in relevant economic evaluations (see Results of the review of economic studies and Overview of other published economic studies of interest).

Base-case results for decision question 1

Base-case results for decision question 2

Scenario analysis: cancer detection rates

Our original base case used RRs of cancer detection estimated from NMA of six RCTs, including the Hara 200826 and Takenaka 200828 trials classified as comparisons between LATP (without freehand device) and LATRUS. There are uncertainties over this approach due to the types of anaesthesia used in the Hara and Takenaka trials (spinal injection in the transperineal arm and caudal block in the transrectal arm). We therefore excluded the Hara and Takenaka trials from the revised base case, as reported above.

Alternative NMA scenarios were reported in Table 31:

1. Hara and Takenaka classified as LATP-any versus LATRUS (for decision question 1) and LATP-other versus LATRUS (for decision question 2).

2. Hara and Takenaka classified as GATP versus LATRUS.

Results of these scenarios for subgroup A decision question 1 are reported in Table 38. For NMA scenario 1, the ICER for LATP versus LATRUS is £28,322 per QALY in subgroup A, increasing to £31,261 per QALY in subgroup D. For NMA scenario 2, the ICER for LATP versus LATRUS is £10,096 per QALY in subgroup A, increasing to £21,322 per QALY in subgroup D. GATP remains dominated in NMA scenarios 1 and 2 for all subgroups.

For decision question 2, results are not sensitive to the NMA scenarios. This is because the RR for cancer detection with LATP-freehand versus LATRUS does not change between NMA scenarios and other comparators are dominated in all scenarios and subgroups (see Table 39 for subgroup A).

We also investigated the effect of using observational data to estimate cancer detection rates: see Appendix 9, Tables 83 and 84. Results were not sensitive to the observational scenarios. For decision question 1, the ICER for LATP-any versus LATRUS was well below £30,000 per QALY for all observational scenarios and subgroups. For decision question 2, the ICER for LATP-freehand versus LATRUS was below £20,000 per QALY for all observational scenarios and subgroups. GATP and LATP-other in decision question 2 had high ICERs or were dominated in all observational analyses.

Scenario analysis: probability of biopsy complications

The rationale for choosing sources for the incidence of biopsy-related complications is explained in Biopsy-related complications above. For the base case, we used admission rates reported by Tamhankar et al.. 80 Additional scenario analyses to test the effect of alternative estimates of biopsy-related admission rates are reported below:

1. Inclusion of additional overnight stays immediately after biopsy as reported by Berry et al.. 94,97 This increases the overall admission rate for transperineal biopsies more than for transrectal biopsies.

2. Inclusion of additional overnight stays from the Berry study applied to GATP only. 97

3. Rosario et al. 95 used as the source of admissions. This reduces the admission rate for LATRUS compared with the base case.

4. Pepe and Aragona96 used as the source of admissions for transperineal biopsies. This reduces the admission rate for LATP and GATP.

5. Rosario study and Pepe and Aragona study as the sources of admissions for LATRUS and transperineal biopsies respectively.

Table 40 (decision question 1) and Table 41 (decision question 2) show the results of these scenarios for subgroup A (MRI Likert score 3+ at first biopsy).

For decision question 1, results are sensitive to the difference in admission rates between LATP and LATRUS. In scenario 1 with additional overnight stays included, the ICER for LATP is over £70,000 per QALY for subgroup A, and higher for other subgroups. In scenario 3, with a lower admission rate for LATRUS, the ICER for LATP versus LATRUS is £17,119 per QALY for subgroup A, and over £30,000 per QALY for other subgroups.

GATP is dominated in all scenarios.

For decision question 2, ICERs for LATP-freehand are higher in scenario 1 with the overnight stay included: £19,140 per QALY for subgroup A and over £30,000 for other subgroups. LATP-other and GATP are dominated in all scenarios.

We note that the excess overnight stays after TP from the Berry et al. study relate to Hospital Episode Statistics from 2014 to 2017, when the use of local anaesthetic for TP was rare. This suggests that scenario 2 may be a more appropriate interpretation of results from the Berry study than scenario 1.

Scenario analysis: cost of core samples

Scenario analysis: biopsy costs

Other scenario analysis

Other scenario analyses are presented in Appendix 9, including the probability of repeat biopsy, and RR of cancer detection from observational studies. Appendix 9, Table 88 presents scenario analyses conducted for decision questions 1 and 2 in subgroup A with a lower impact in the model results and that did not impact the final conclusions. The results for the other subgroups (B, C and D) follow the same tendency as the results presented in Appendix 9, Table 88.

Generalisability

The TP protocols (e.g. device used/sampling method/number of cores taken) varied between studies, which may partly reflect local clinical practice guidelines in study host institutions, but also the evolution of transperineal prostate biopsy practices over time (e.g. increases in the number of cores sampled over time as protocols evolve). Some of the more recently published studies used pre-biopsy mpMRI to inform biopsy sampling, but this constitutes a small proportion of the whole evidence base. This presents a challenge to the applicability of the evidence to UK clinical practice, where mpMRI is recommended for routine use to inform the decision to refer for biopsy and may also be used to target biopsy sampling.

The studies were typically conducted in single centres by clinical investigators using local biopsy protocols to evaluate the optimum biopsy modality in their centre, on a range of outcomes such as use of general or local anaesthesia protocols, procedure time and related resources, biopsy complications and the patient’s ability to tolerate pain and discomfort during and after the biopsy. Many studies pre-date the introduction of mpMRI into prostate biopsy protocols, and given the preponderance of studies from East Asia use of mpMRI worldwide may differ from practice in the UK.

The multicentre UK study (TRANSLATE52–54) will provide evidence for freehand LATP using any ultrasound probe-mounted needle guidance device, including the PrecisionPoint and UA1232 devices. As the study uses freehand devices to perform the biopsies, it is expected to inform any future consideration of both decision question 1 (LATP-any vs. LATRUS) and decision question 2 (LATP-freehand vs. LATRUS). This will be the first available comparative evidence for the UA1232 device, and is expected to provide information on cancer detection, infection rates and other outcomes including cost.

Strengths

We conducted a systematic review of evidence related to the decision questions specified in the NICE scope, with pairwise and NMA of cancer detection outcomes from both randomised and observational studies.

A major strength of the economic analysis is that we could build on the work of previous researchers to develop an appropriate decision-tree structure and model parameters, including the economic evaluation of the PROMIS study by Faria et al., the adaptation of the PROMIS analysis by Wilson et al. and the economic model that informed the update of the NICE guideline (NG131). 63,65,67,115 The decision tree is based on prevalence and diagnostic yield data for TRUS from the PROMIS study, which used estimates of true disease status based on a template mapping biopsy as the reference standard.

Another strength is that the predicted impact of diagnostic yield on long-term costs and outcomes was based on the recent and high-quality economic model that was developed to inform an update of the NICE guideline for prostate cancer (NG131). The NICE economic model has gone through a rigorous process of development, review and discussion by members of the guideline committee (including topic specialists and methodological, patient and public experts) and consultation with stakeholders. We appreciate that the NICE Centre for Guidelines provided a copy of this model, as this helped us to replicate the transition probabilities accurately (in particular it provided access to the covariance matrices for the calibrated parameters).

The RR of cancer detection was directly informed by the clinical effectiveness systematic review and therefore we believe that the most relevant studies reporting data on cancer detection rates were considered.

Limitations

The clinical evidence base and economic model have several limitations. As discussed above, there are limitations to the generalisability of the clinical evidence to UK practice, including variations between centres and over time in TP protocols and the use of mpMRI to inform referral for biopsy, and biopsy sampling.

The definition of patient subgroups in the model was based on mpMRI Likert scores, in order to align with epidemiological data from the PROMIS study. However, we are aware that some UK centres use the PI-RADS method to summarise mpMRI results. We have not provided results for subgroups according to lesion site or prostate volume, due to lack of data to differentiate prognosis or diagnostic yield of the biopsy methods under assessment.

We extrapolated data on repeat biopsy from LATRUS and GATP (based on the Jimenez et al.’ study) to LATP, in the absence of specific evidence for LATP. 113 Moreover, the Jimenez et al. study assesses a Spanish cohort that may not be wholly generalisable to UK practice. However, a scenario analysis on the probability of repeat biopsy showed that the model results are not sensitive to this parameter.

We have assumed that patients with a negative biopsy result were discharged and no additional costs were incurred, since we are uncertain about the extent and nature of the follow-up of these patients in primary care. However, it is likely that a substantial proportion of people with a negative biopsy who develop prostate cancer later have a diagnosis based on symptoms, which is considered in the model. Lastly, although we included costs for recently recommended treatments for metastatic hormone-sensitive prostate cancer (apalutamide and enzalutamide), we did not incorporate survival benefit from these treatments in our model. Our scenario analysis showed that excluding apalutamide and enzalutamide from the treatment options for mHSPC has a low impact in the model results.