Photodynamic versus white-light-guided resection of first-diagnosis non-muscle-invasive bladder cancer: PHOTO RCT

Incidence

Bladder cancer is the most frequently occurring tumour of the urinary system, with > 10,500 new cases diagnosed each year in the UK. 2 Overall, bladder cancer is the 11th most common cancer in the UK, accounting for 3% of all new cancer cases. 2 Histologically, > 90% of bladder cancers are of the transitional cell carcinoma type. Bladder cancer is more common in men than in women (5 : 2 ratio), making it the eighth most common cancer in men and the 16th most common cancer in women. 2 Incidence rates for bladder cancer in the UK are highest in people aged 85–89 years, with 8 in 10 cases occurring in people aged ≥ 65 years. Cigarette smoking is causally related to over one-third of bladder cancer diagnoses and is also a risk factor for progression to cancer-related death. 3,4 Time trends in bladder cancer incidence rates over the past 10–20 years are difficult to interpret because of changes in classification. There is a trend towards a reduction in age-standardised incidence rates, currently 11 cases per 100,000 population, which is predicted to continue to decline at a rate of 1% annually. 5 However, the growth in the ageing population will have a substantial impact on the total number of cases, which is projected to rise at an annual rate of > 1%. 5

Pathology

The extent of the spread of cancer is described using the American Joint Committee on Cancer (AJCC) and the International Union Against Cancer (UICC) tumour node metastasis (TNM) staging system. 6 Tumours confined to the epithelial lining (urothelium) are classified as stage Ta and those invading the lamina propria are classified as stage T1 (Figure 1). Ta and T1 tumours can be easily removed through transurethral resection (TUR) and, therefore, for therapeutic purposes, are grouped together as non-muscle-invasive bladder cancer (NMIBC). NMIBC also includes flat, high-grade tumours that are confined to the epithelium, which are classified as carcinoma in situ (CIS). Grade (microscopic characteristics of the tumour cells) can be used to describe the aggressiveness of cancers, which are characterised as either low grade (relatively benign) or high grade (aggressive). 6

FIGURE 1.

Tumour node metastasis staging of bladder cancer (AJCC TNM staging system). The primary tumour (T) stage shown in this figure is CIS. Node (N) stage and metastasis (M) stage measure cancer spread to local lymph nodes and other parts of the body, respectively.

Presentation and diagnosis of bladder cancer

The most common presentation of bladder cancer is haematuria (presence of blood in the urine), which may be associated with additional symptoms such as dysuria (painful urination), increased frequency/urgency of urination, failed attempts to urinate or urinary tract infections. Haematuria is either visible (frankly blood-stained urine) or non-visible (urine that is clear-looking to the naked eye). Non-visible haematuria is detected by dipstick or microscopic examinations, often included in standard primary care assessments for an NHS Health Check or in the investigation of urinary symptoms. Bladder cancer is detected in approximately 10% of patients with visible haematuria and 3–5% of those with dipstick or microscopic haematuria who are aged > 40 years. 7,8 Therefore, these patients are urgently referred for assessment in rapid-access haematuria clinics in secondary care, where suspected bladder tumours are usually detected by cystoscopy under local anaesthetic or, less frequently, on imaging by ultrasound scanning or computerised tomography (CT). Visual appearances of bladder cancer are confirmed by histological diagnosis using samples taken during cystoscopic transurethral resection of bladder tumour (TURBT), which is conducted as part of the initial management of bladder cancer.

Initial management of non-muscle-invasive bladder cancer

About 80% of people with a new diagnosis of bladder cancer will have NMIBC and will initially be treated by TURBT. The subsequent goal in NMIBC management is the prevention of recurrence and/or progression to higher-stage, life-threatening, muscle-invasive disease. It is thought that failure to identify/appreciate tumours is a factor in 20–40% of the recurrent tumours that are overlooked. 9,10 Tumour seeding following resection and urothelium that may be genetically ‘primed’ for new tumours developing are other factors that are considered relevant and will have an impact on recurrence rates independent of the completeness of resection. Recurrence and stage progression to muscle-invasive or metastatic cancer is more likely to occur in those with high-grade tumours with concomitant CIS. In particular, CIS, which is a flat tumour, can be easily missed using conventional white-light (WL) resection. 11

Risk of recurrences and stage progression

Both clinical and histological parameters can be used to estimate individual risk of recurrence and progression of NMIBC into muscle-invasive bladder cancer (MIBC). Based on this, the European Organisation for Research and Treatment of Cancer (EORTC) Genitourinary Group has developed an algorithm that calculates probabilities of recurrence and progression, which are integral to the current European Association of Urology (EAU) guideline on the treatment of NMIBC. 12

These probabilities are based on the number of tumours, tumour size, prior recurrence, histological T stage, presence of CIS and tumour grade. The risks of recurrence and progression 3 years after diagnosis are summarised in Table 1. Patients’ cancer management plans are tailored to the risk categories in terms of the intensity of follow-up and the use of adjuvant therapies.

Strategies to reduce recurrence and progression

Health economic impact of managing non-muscle-invasive bladder cancer

Non-muscle-invasive bladder cancer is one of the costliest cancers to manage because of its high prevalence and high recurrence rate and the need for adjuvant treatments and long-term cystoscopic surveillance. The total cost of treatment and 5-year follow-up of UK patients with NMIBC increased from £73M to £213M from 2001 to 2012 (inflation corrected). This was due to an ageing population and better-defined adjuvant treatment and surveillance regimens requiring additional therapies, with potential mortality and long-term morbidity (e.g. radical surgery). 20,21 Health-related quality of life (HRQoL) is also known to be affected in those receiving morbid treatments for bladder cancer. 22 The cost-effectiveness of NMIBC treatment strategies has not been widely studied.

Health need

Although many cases of NMIBC are readily treatable with cystoscopic resection, it remains a major health-care burden, for the reasons described above. 11,20 From a patient perspective, there are often considerable anxieties about recurrences, TUR and the impact of adjuvant treatments. 23 TUR is associated with reduced QoL across both mental and physical health domains, although these effects are usually transient. 24 Substantial effects on HRQoL are most likely to result from adjuvant intravesical treatments and radical or palliative treatments for progression. 22 More efficient management strategies to reduce NMIBC recurrence, and hence decrease both the burden to patients and costs to the NHS, are urgently needed. One such approach currently available in the NHS is photodynamic diagnosis (PDD).

Photodynamic diagnosis of non-muscle-invasive bladder cancer

Primary objectives

Secondary objectives

Additional objectives

• Lay the basis for modelling the safest and most cost-effective cystoscopic follow-up surveillance schedules.

• Evaluate the learning curve for the procedure to account for its effects on outcomes of both PDD and standard WL resections.

• Establish a well-characterised cohort of patients with intermediate- and high-risk NMIBC, including clinical data and urine, blood and tumour specimens that would be available for separately funded research of genotypic and phenotypic studies.

Eligibility criteria

Recruitment procedure

An eligibility checklist was completed by the local PI (or a delegate) to assess fulfilment of the entry criteria for all patients considered for the study. Information from the diagnostic cystoscopy was used to assess eligibility.

Informed consent

All potentially eligible patients were provided with an information sheet to explain why they had been approached and the nature of the study. Eligible patients were asked to provide written informed consent to take part in the study only after they had had sufficient time to consider the trial and the opportunity to have any further questions addressed by the local clinical team.

Technique of photodynamic diagnosis

Photodynamic diagnosis requires the preliminary instillation of the photosensitiser HAL (85 mg in 50 ml of phosphate-buffered saline) into the participant’s bladder through a urethral catheter. Participants were asked not to pass urine for at least 1 hour after instillation. Following operating theatre preparation in accordance with local standard procedures and under appropriate anaesthesia, participants underwent TURBT of their bladder tumour under blue-light illumination of the bladder (wavelength 380–450 nm). A specialised light source, cystoscope, light cables and cameras were required. When using PDD, normal bladder epithelium appears blue, and red areas are considered suspicious and should be resected.

Technique of standard white-light cystoscopy

The control group did not have any preliminary photosensitiser instillation, and standard tumour localisation and resection took place under WL illumination of the bladder (wavelength 400–800 nm).

Second resection

If, in accordance with the EAU guideline,12 the local PI deemed a second TURBT was required, the same method (PDD or WL) was used as that of the participant’s trial treatment allocation.

Adjuvant therapy

Adjuvant therapy was prescribed according to local clinical judgement in accordance with participant characteristics and the EAU guideline. 12

Treatment allocation (randomisation concealment and blinding)

Eligible consenting patients were centrally randomised using either the secure web-based randomisation system or the 24-hour interactive voice-response randomisation system at the Centre for Healthcare Randomised Trials (CHaRT) in Aberdeen. Minimisation by centre and sex was used to allocate participants 1 : 1 between the control (WL) and experimental (PDD) groups. The minimisation algorithm incorporated a random element to prevent potential deterministic treatment allocation. Participants were randomised by clinical teams at the participating NHS hospitals.

Outcome measures

The primary clinical outcome measure was the time to recurrence measured as time from the day of randomisation to the day of subsequent biopsy for pathologically proven first recurrence (including progression, cystectomy and death from bladder cancer). The primary objective was to compare the time to recurrence for the two treatment strategies, with a principal point of interest at 3 years.

The primary health economic outcome was to evaluate cost-effectiveness using the incremental cost per recurrence avoided and cost–utility as the incremental cost per QALY gained at 3 years.

Secondary outcome measures

Other clinical outcomes included adverse events (AEs) and complications up to 3 months from initial or second TURBT (as applicable). Direct, surgically related, postoperative events occurring within the 30 days following TURBT were assessed using the Clavien–Dindo classification for surgical complications. 31 Events occurring up to 3 months after TURBT were assessed and recorded using the National Cancer Institute Common Terminology Criteria for Adverse Events (CTCAE) v4.0 framework. 32

The relative changes in HRQoL resulting from the physical and psychological benefit, together with any harms associated with each strategy and any subsequent necessary cancer treatment, were measured using the generic EuroQol-5 Dimensions, three-level version (EQ-5D-3L), questionnaire; the cancer-specific European Organisation for Research and Treatment of Cancer – Quality of Life Questionnaire – Core 30 (EORTC-QLQ-C30); and the disease-specific European Organisation for Research and Treatment of Cancer – Quality of Life Questionnaire – Non-Muscle Invasive Bladder Cancer – 24 items (EORTC-QLQ-NMIBC-24) questionnaire. These were completed by participants on paper at baseline (prior to knowledge of treatment allocation), following surgery and at 3, 6, 12, 18, 24 and 36 months after randomisation.

Disease progression was defined as an increase in stage to MIBC or the development of nodal or metastatic disease. In addition, details of intravesical treatment failure (e.g. BCG) were captured. Overall survival and bladder-cancer-specific survival were compared between the two treatment groups.

Other cost-effectiveness outcomes included the estimation of the incremental cost per recurrence avoided using the economic model over the patients’ lifetime and the estimation of the incremental cost per QALY gained using the economic model over the patients’ lifetime.

Data collection and management

The PHOTO trial schedule of assessment and investigations is summarised in Table 2. Eligibility was checked during routine attendance for diagnosis and staging of new bladder cancers, which included obtaining medical history. Eligible patients who consented to join the trial completed HRQoL questionnaires prior to initial TURBT and again prior to discharge from hospital. Intraoperative and postoperative data were reported by the local research teams at the time of the initial and second (as applicable) TURBT. At first recurrence, the date of biopsy and confirmatory histological details were reported by the local team. Details of any cystoscopy checks that were carried out were reported by the local research teams from the routine clinic visits at 3, 6, 9, 12, 18, 24 and 36 months post initial/second TURBT (as applicable), including an assessment of AEs at 3 months. Disease progression was assessed by the local research team using the results of further resection or imaging during follow-up. A short case report form (CRF) was completed annually from routine urology clinic visits following recurrence or progression to collect the disease status (e.g. further recurrence of NMIBC, progression to MIBC or cystectomy) and survival status of participants. The outcome of these assessments for each participant were entered as appropriate by centre staff on electronic CRFs on the central secure database held by CHaRT, where the accruing data were monitored.

Costs and changes in HRQoL were collected via self-completed postal questionnaires sent directly to participants by CHaRT at 3, 6, 12, 18, 24 and 36 months post randomisation. In addition, a participant costs questionnaire was administered by post at 30 months post randomisation.

All recruiting surgeons completed a learning curve questionnaire to elicit their WL- and PDD-resection experience prior to their centre commencing recruitment. The subsequent accruing experience of each surgeon was captured on CRFs.

All CRFs and participant questionnaires are available on the NIHR Journals Library project web page (www.journalslibrary.nihr.ac.uk/programmes/hta/1114202/#/documentation; accessed 16 November 2021).

Tracking and monitoring adverse events

The CHaRT study office was notified of AEs (CTCAEs, primarily grade ≥ 3) by the local research team on the 3-month cystoscopy CRF. Unrelated AEs were not recorded. In the PHOTO trial, ‘relatedness’ was defined as any untoward medical event that had a reasonable causal relationship with PDD-TURBT, standard WL-TURBT or the intravesical MMC. The following events were potentially expected: anaemia, bladder discomfort/pain, bladder perforation, bleeding resulting in clot retention, constipation, diarrhoea, deep-vein thrombosis (DVT), fever, gout, haematuria, headache, increase in white blood cell count, increased level of bilirubin, insomnia, nausea, postoperative dysuria, prolonged catheterisation, skin rash, ureteric obstruction/hydronephrosis, urethral stricture, urinary frequency, urinary retention, urinary tract infection and vomiting.

Any serious adverse events (SAEs) related to the participants’ TURBT treatment that were not further interventions (e.g. being admitted to hospital for an infection) were recorded on the SAE form. All deaths from any cause (related or otherwise) were recorded on the SAE form.

It was a requirement to report to the sponsor any SAEs that were deemed related and unexpected within 24 hours of receiving the signed SAE notification. Such SAEs were also reported to the main REC within 15 days of the chief investigator becoming aware of the event. All related SAEs were summarised and reported to the appropriate authorities as required.

Trial oversight

The day-to-day management of the trial was provided by the Clinical Trials and Statistics Unit at the Institute of Cancer Research (ICR-CTSU) and CHaRT based within the Health Services Research Unit, University of Aberdeen, with ICR-CTSU leading on trial management and CHaRT coordinating data management and statistics. The trial offices provided day-to-day support for the PIs at recruiting centres. The PIs, supported by dedicated research nurses, were responsible for all aspects of local organisation, including recruitment of participants, delivery of the interventions and notification of any problems or unexpected developments during the study.

A core group, chaired by the chief investigator and comprising key members of CHaRT and ICR-CTSU as well as the Health Economics and Biorepository teams at Newcastle University, met monthly. A Trial Management Group (TMG) was established and included the chief investigator, scientific leads (from ICR-CTSU and CHaRT), a health economist, co-investigators and collaborators, a patient representative, CHaRT’s Senior information technology manager, a trial statistician, and senior project managers and trial managers from ICR-CTSU and CHaRT. PIs and other key study personnel were invited to join the TMG, as appropriate, to ensure that there was representation from a range of centres and professional groups. The TMG met regularly and at least annually, and had operational responsibility for the conduct of the trial. The TMG’s terms of reference, roles and responsibilities were defined in a charter issued by ICR-CTSU. The trial was further overseen by the independent Trial Steering Committee (TSC) and an independent Data Monitoring Committee (DMC). The TSC comprised an independent chairperson and four independent members. The DMC comprised an independent chairperson and two independent members.

Patient and public involvement

A co-investigator (a patient with bladder cancer) provided advice based on the service user perspective and contributed to user-led selection of patient-reported outcome measures (PROMs). His experiences of the diagnosis and treatment of bladder cancer were taken into consideration in the design of this trial; in particular, this helped inform the use of QoL questionnaires that included emotional, social and physical domains specific to bladder cancer. As a member of the TMG, he also advised on approaches to recruitment, participant information resources and the dissemination of findings.

One of the independent members of the TSC was also a patient representative. The TSC met throughout the trial and reviewed all study documentation, including patient-facing documents. In addition to being an integral part of the study oversight and contributing to the Plain English summary included in this report, he provided feedback on what he felt were the key impacts of the study and the value of his contributions as the patient and public involvement (PPI) representative on the independent TSC:

As a patient who has lived experience of both a white light and a blue light, PDD-TURBT, I was particularly interested in both the outcome and the conduct of this trial from a patient perspective. Equally, in my role as Chair of the charity Action Bladder Cancer UK, I was keen to support much needed research into bladder cancer, and to use social media and speaking opportunities to raise the profile of the trial and promote participant recruitment. I have supported the TSC since its inception in 2014, being an active part of the committee and helping to make the reporting more patient focused. I was alert to recruitment issues early on (common to all bladder cancer trials) and able to ensure that the TMG, including its own PPI, was doing all it could to improve recruitment rates, paying particular attention to patient-facing materials. Overall, the trial has been very well run.

PPI TSC member

Sample size

The trial aimed to detect an absolute reduction in recurrence of 12% at 3 years: from 40% (under the conservative assumption that all of the patients recruited would be intermediate-risk patients with a probability of recurrence of 0.4 at 3 years) to 28% (similar effect sizes of photodynamic therapy have been reported in both intermediate- and high-risk groups27). This is equivalent to a relative reduction of 30%. Power calculations were based on log-rank analysis of time-to-event data, translating an improvement in the fixed time point recurrence-free rate of 60–72% to a target effect size hazard ratio (HR) of 0.64. The recruitment of 533 participants (214 recurrences) would enable the detection of a HR of 0.64 between experimental and control strategies, and would, using the log-rank test, provide 90% power at a two-sided 5% significance level. This calculation assumed 2.5 years of staggered recruitment (with 6%, 13%, 21%, 29% and 31% of the total number of patients recruited within each 6-month period), a minimum of 3 years’ follow-up and cumulative follow-up attrition rates of 0.56% by the end of year 1, 1% at the end of year 2 and 6.4% at end of year 3, based on unpublished data from the Bladder cyclooxygenase 2 Inhibition Trial (BOXIT).

To achieve the recruitment target, we planned to activate 30 secondary care sites that expected to see approximately 4590 new bladder cancer diagnoses over 2.5 years, from which we would exclude patients with MIBC (20%) and, from the remaining NMIBCs, those with low-risk disease (50%). Furthermore, we predicted that only 30% of these patients would be recruited based on willingness to participate or missed opportunities for recruitment.

Statistical methods

The trial analysis followed a statistical analysis plan. There was no interim analysis of clinical effectiveness data, only a single analysis after the database was locked on 2 June 2020. The main analyses for both safety and clinical effectiveness were based on the intention-to-treat (ITT) principle, that is they were analysed as randomised, regardless of the intervention received. However, after the initial resection, some participants were found to have no tumour, some were found to have MIBC and some had an early cystectomy. These participants were excluded from the ITT population (see Chapter 3 for details). We also used per-protocol analyses restricted to participants who received the treatment to which they were allocated.

Baseline and follow-up data were summarised using appropriate statistics and graphical summaries. All analysis was done using Stata, version 16 (StataCorp LP, College Station, TX, USA), software.

Primary outcome

The primary outcome was time to recurrence of bladder cancer, measured in months from randomisation to recurrence, progression, cystectomy or death due to bladder cancer. The principal time point of interest was 3 years. We used a time-to-event framework to analyse the primary outcome. In the primary analysis, deaths were treated as censored in Cox proportional hazards regression models. The first model adjusted for the minimisation covariate factors of sex and centre (the latter via a random-effects frailty model) and the treatment effect was summarised as a HR with a 95% CI. The second model added the following known prognostic factors: smoking status, risk group, presence or absence of CIS and grade of surgeon (i.e. registrar, non-consultant career grade or consultant). The HRs and 95% CIs for these prognostic factors were given, as was the HR for the treatment effect adjusting for these prognostic factors.

We plotted empirical survival distribution using Kaplan–Meier plots; visually, the proportional hazards assumption was questionable. The proportional hazard assumption was assessed by plotting log [−log (survival)] against log (analysis time). We used accelerated failure time models based on Weibull, exponential, log-logistics, log-normal and generalised gamma distributions, relaxing the proportional hazards assumption. The model with the smallest Akaike information criterion value was considered the best-fitted model and reported.

Secondary outcomes

Secondary outcomes were analysed using the appropriate generalised linear models as follows:

• Disease progression at 3 years – progression was defined as increased stage to MIBC, development of metastatic disease at other regional sites, development of nodal disease or death due to bladder cancer. The end point was analysed using a time-to-event framework. Analysis was as for the primary outcome.

• HRQoL – standard measure-specific algorithms were used to derive scores from and handle missing data within each HRQoL outcome. We used a linear mixed model (random effect for centre and participant; fixed effect for nominal time, treatment, sex, smoking status, risk group, presence or absence of CIS and grade of surgeon) to analyse the repeated measures of HRQoL outcomes. Treatment effects at each time were derived from the interaction term for time by treatment. To account for multiple HRQoL outcomes and time points, we report 99% CI around estimates.

• Bladder-cancer-specific survival – the time to bladder-cancer-specific death was analysed using a competing risks approach (based on the Fine and Gray model). 33 Death from other causes was considered a competing risk in the Cox proportional hazards model, instead of assuming non-informative censoring. The first model was adjusted for sex and centre, and the second was adjusted for sex, centre, smoking status, risk group, CIS and surgeon grade.

Sensitivity analysis

A sensitivity analysis of the primary outcome that treated deaths from non-bladder cancer causes as a competing risk rather than non-informative censoring was performed. We report the subhazard ratio (SHR) for the treatment effect in two models, first adjusted for sex and centre, and then adjusted for sex, centre and prognostic factors as per the primary outcome. We plotted cumulative incidence curves for time to recurrence.

Surgical learning curve

The effect of photodynamic resection experience (learning curve) on clinical effectiveness was examined. Based on the results of a questionnaire completed by participating surgeons prior to centre activation, each centre was classified as PDD experienced or naive.

A subgroup analysis comparing outcomes from experienced or naive centres, including specific PDD- and WL-resection-related outcomes, was used to assess the maximum effect of experience on outcome in an NHS setting. Early recurrence (at 12 weeks) was used as a proxy of incomplete resection. The subsequent accruing experience of each surgeon was captured on a CRF. This allowed each randomised participant to be positioned on an individual surgeon learning curve.

Safety data

The proportion of participants experiencing AEs (CTCAE grade 3 or above) was compared between groups using Poisson regression. The number of AEs by Clavien–Dindo grade was tabulated by group.

Health economics methods

Economic evaluations are conducted as an aid to decision-making. In the PHOTO trial, the economic evaluation aimed to determine whether or not PDD-TURBT was cost-effective compared with WL-TURBT as part of the management of people with suspected intermediate- and high-risk cancers confined to the bladder lining. For the economic evaluation, both a trial-based and model-based cost–utility analysis was planned. The within-trial analysis considered costs and outcomes over a 3-year follow-up. The model-based analysis was planned to have a lifetime time horizon. The model-based analysis was not conducted as it was judged not to add any additional information for decision-makers. See Appendix 2 for information about the proposed modelling component.

The within-trial economic evaluation took an NHS and Personal Social Services (PSS) perspective in line with the NICE reference case. 34 The results were estimated at 3 years following randomisation using the ITT principle. Results are presented in terms of cost, QALYs, incremental cost, incremental QALYs and incremental cost per QALY. The costs and QALYs accruing after the first year and the incremental cost per QALY gained were discounted at a rate of 3.5% per annum.

A time trade-off (TTO) study was commenced to provide estimates of short-term decreases in utility following cystectomy; these data were planned to be used alongside the EQ-5D-3L trial data collected at 6-month intervals. However, the TTO work was paused because of COVID-19 and the nature of the interviews (involving complex instructions, face-to-face contact and lasting approximately 1 hour); it was not feasible to continue this work within COVID-19 restrictions. Eleven patients were recruited, which was substantially short of our target sample size (n = 50) and, hence, meaningful analysis could not be performed. See Appendix 2 for a brief description of the method.

Incremental costs and QALYs were presented as point estimates and 95% CIs using adjusted linear regression. The on-parametric bootstrap approach was also used to produce the cost-effectiveness plane, representing the uncertainty in incremental cost and QALY estimates, and the cost-effectiveness acceptability curve (CEAC), representing the probability that PDD-TURBT was cost-effective compared with WL-TURBT at different cost-effectiveness thresholds for an incremental cost per QALY gained. 35

The within-trial analysis was conducted in Stata version 16.1. The model-based analysis was to be conducted in TreeAge Pro™ 2020 (TreeAge Software, Inc., Williamstown, MA, USA).

Initial procedure

The resources associated with the initial procedure included all of the resources incurred until discharge. The resource use data required to deliver each intervention were collected prospectively for every participant in the study. The operative details were recorded at the time of surgery (e.g. time in theatre, grade of operating surgeon). These data were captured in the operation details CRF. Costs incurred after the TURBT procedure but before discharge were collected using the initial resection CRF and post-treatment participant questionnaire. These forms contained information on the length of hospital stay for the initial TURBT (based on admission and discharge dates), medical procedures and medical events that could occur during the treatment phase.

Subsequent use of services following discharge for the index procedure

After participants were discharged, their resource use was captured using the health service utilisation questionnaire (HSUQ) completed by participants at 3, 6, 12, 18, 24 and 36 months. The use of the HSUQ allowed us to categorise resource use as either secondary or primary care. This included all secondary care (e.g. WLC, flexible cystoscopy, mitomycin, BCG, CT scan, cystectomy, palliative care, inpatient admissions, day admissions, hospital doctor consultation, outpatient consultations, accident and emergency consultations) and primary care contacts with health professionals (e.g. GP consultations, practice and district nurse consultations, other health professional consultations). Visits with these health professionals could occur at the health-care practice, at the participant’s home or over the telephone. We distinguished between the different types of consultation to account for the different costs associated with each setting.

For our analysis, we assumed that any participant who partially completed the questionnaire had left other responses blank because the questions did not apply to them. For participants who had died during the follow-up period, their resource use was automatically imputed as 0. This could cause an underestimation in our costs as these participants may have used some services during the data collection period before they died.

Unit costs of NHS care

National average unit costs were applied to resource use data to generate the total costs to the health service. The sources of unit costs were the British National Formulary (BNF)36 and the NHS Reference Costs 2018–1937 for secondary care resource use data, and the Personal Social Services Research Unit (PSSRU)’s Unit Costs of Health and Social Care for primary care resource use data. 38 See Appendix 2, Table 46, for a list of all unit costs used in the within-trial economic analysis, together with their sources. These costs were reported in 2018–19 Great British pounds (£). For the purpose of inflation, we utilised the Campbell and Cochrane Economics Methods Group Evidence for Policy and Practice Information Centre (CCEMG-EPPI-Centre) Cost Converter, version 1.6 (Campbell and Cochrane Economic Methods Group, London, UK), using International Monetary Fund-reported inflation data. 39

Estimation of NHS costs

For each participant, the total use for each resource was multiplied by the unit cost to calculate the total cost of each resource. For example, the initial length of admission was multiplied by the NHS cost per night for an inpatient stay on a general ward to obtain the cost of hospitalisation for each participant. The cost for each year beyond the first year was discounted at a rate of 3.5% per annum. The total discounted costs from the health services perspective were calculated by summing all intervention treatment and follow-up discounted costs for each participant in the data set.

Participant- and companion-incurred costs and indirect costs

Participant resource utilisation comprised three main elements:

1. costs of accessing and using NHS health services (e.g. petrol, public transport and parking)

2. time costs of accessing and using NHS health services (e.g. time involved away from usual activities or work)

3. indirect costs due to ill health.

Costs of accessing and using NHS health services

The estimation of costs of accessing and using NHS health services required information from participants about the number of visits to, for example, their GP (estimated from health-care utilisation questions in the participant costs questionnaire administered at 30 months) and the unit cost of making a return journey to each type of health-care provider.

Time-off costs of accessing and using NHS health services

The cost of participant time was estimated in a similar manner. The participant was asked, in the participant costs questionnaire, how long they had spent travelling and attending their last visit to each type of health-care provider and if they had been accompanied by a friend or relative (if so, the companion’s time and travel costs were also incorporated into the analysis). These data were recorded in their natural units (e.g. minutes). The unit cost of time lost was obtained from the Department for Transport estimates for the value of work and leisure time. 40 The cost of each visit was then calculated by multiplying the time lost by the time unit cost. The total time cost per patient was then calculated by multiplying the patient’s time cost per visit by the number of health-care contacts obtained from the health-care utilisation questions.

Indirect costs due to ill health

Indirect costs were defined as the production losses when the participant was unable to return to work or was required to take sick leave because of their illness. The cost of days lost was estimated using the UK median gross hourly wage. When a participant’s self-reported costs associated with a specific type of health service visit were missing, the mean cost for that type of visit was imputed. Participants completing the participant costs questionnaire were asked how many days they had been off work in the previous 2 months as a result of health problems. These data were collected using the participant costs questionnaire at the 30-month follow-up point. The data were recorded in their natural units (e.g. days) and multiplied using the unit costs. The total production losses due to time away from work as a result of health problems were estimated and compared between treatment groups.

The data on costs and time-off costs of accessing and using NHS health services and the indirect costs due to ill health were summed to generate a total cost for each participant and their companions. The incremental cost differences between groups from a participant perspective were estimated using the same methods outlined for the NHS perspective (see Health economics methods).

Quality-adjusted life-years

Participants were asked to complete the EQ-5D-3L instrument41 at baseline, discharge and follow-up visits (3, 6, 12, 18, 24 and 36 months). The EQ-5D-3L is a generic instrument used to assess participants’ QoL for the base-case analysis because it is the preferred utility measure of NICE. 34 The EQ-5D-3L measure divides health status into five dimensions (mobility, self-care, usual activities, pain/discomfort and anxiety/depression). Each of these dimensions has three levels, so 243 possible combinations of health states exist. Each combination of levels across the dimensions is associated with an EQ-5D-3L index value. 41 Utility value data derived from the EQ-5D-3L were combined with mortality data from the trial under the standard assumption that all patients who have died in the trial will have a utility value of 0 from the date of death to the end of follow-up. The QALY for each year was then calculated based on these assumptions using the area-under-the-curve approach, and assuming linear extrapolation of utility between time points. QALYs for each year beyond the first year were discounted at a rate of 3.5% per annum. The total discounted QALYs for each participant were calculated by summing the discounted QALYs over the trial follow-up period.

Handling missing data

Missing data are a concern in this study because costs or health outcomes in individuals with missing data may be systematically different from those in individuals with no missing data. A substantial proportion of missing data observed in a trial can pose significant problems for data analysis.

The complete-case analysis is inefficient for the PHOTO study because all of the information from individuals with at least one assessment missing is discarded. In addition, the complete-case analysis cannot be considered an ITT analysis because some randomised patients with follow-up data are excluded. 42 Therefore, the imputed data analysis was used as the base-case analysis and the complete-case analysis was conducted as a scenario analysis in the sensitivity analysis. The imputation was undertaken using Stata’s multiple imputation (MI) procedure.

Multiple imputation was used to impute missing EQ-5D-3L utility values and cost values for individuals with data at baseline or at least one follow-up visit. When missing data are ‘missing at random’ (MAR), valid conclusions can be drawn from the available data using the MI approach. 43 Missing values of total follow-up cost and EQ-5D-3L utility values at each time point were imputed using predictive mean matching by treatment allocation group, accounting for the three closest estimates in terms of baseline EORTC recurrence risk group, age at randomisation and sex. Chained equations were used for the imputations. The imputation procedure predicted 50 plausible alternative imputed data sets, which was found to be a sufficient number to provide stable estimates. An analysis of incremental costs and QALYs was undertaken across the 50 imputed data sets and combined to generate one imputed estimate of incremental costs and QALYs. We drew bootstrap samples from each of the 50 imputed data sets.

Estimation of cost-effectiveness

A seemingly unrelated regression (SUR) approach was used to simultaneously estimate the total discounted costs at 3 years and the total discounted QALYs at 3 years, allowing for the likely correlation of costs and effects. 44 For the QALY outcome variables, baseline EORTC recurrence risk group, age at randomisation, sex and baseline EQ-5D-3L utility value were included as covariates. For the cost outcome variables, baseline EORTC recurrence risk group was included as a covariate.

The results are reported as incremental cost per QALY gained for PDD-TURBT relative to WL-TURBT. The incremental cost per QALY was calculated from the coefficient of treatment effect on costs divided by the coefficient of treatment effect on QALYs from the SUR model.

To address the issue of sampling uncertainty in the data, we used non-parametric bootstrapping methods to estimate 95% CIs for the treatment effects on costs and QALYs, using 2000 repetitions. 45 This imprecision was then presented graphically as a cost-effectiveness plane (see Figure 14). This shows the scatterplot of bootstrapped repetitions for incremental costs and incremental QALY pairs for PDD-TURBT compared with WL-TURBT. 46 The scatterplot is divided into four quadrants, each of which represents one of the following scenarios:

1. PDD-TURBT is less costly and more effective than WL-TURBT.

2. PDD-TURBT is more costly and less effective than WL-TURBT.

3. PDD-TURBT is less costly and less effective than WL-TURBT.

4. PDD-TURBT is more costly and more effective than WL-TURBT.

The proportion of the total bootstrap samples that lie in a quadrant represents the probability associated with that scenario.

The bootstrapped estimates of costs and QALYs were also used to produce CEACs. 47 CEACs were generated using the net monetary benefit (NMB) approach, where:

‘QALY’ and ‘cost’ are the estimated total QALYs and total costs for a treatment strategy, respectively, and λ is the decision-maker’s cost-effectiveness threshold for a QALY gained.

In this analysis, λ was varied over the range £0–60,000. In the tabular presentation of the analysis, λ is presented at the following thresholds: £0, £20,000, £30,000 and £50,000.

The proportion of bootstrap samples in which the net benefit is positive at a given threshold for cost per QALY represents the probability that the treatment is cost-effective. This was repeated for each MI data set. The probability across all MI data sets was averaged for the threshold values stated above.

Sensitivity analysis

The base-case analysis was conducted under the MAR assumption, using MI to impute the missing cost and HRQoL values. It is, however, recognised that participants who failed to complete an EQ-5D-3L questionnaire at a specific follow-up assessment may have been in relatively poorer health than those who did. This means that the chance of observing HRQoL could depend on their actual utility value, that is data are likely to be missing not at random (MNAR). Therefore, it is important to explore the impact of the missing data mechanism [i.e. missing completely at random (MCAR), MAR and MNAR assumptions] on cost-effectiveness outcomes. A sensitivity analysis was conducted to explore the impact on the results of assuming that the data are MNAR, with scenarios for systematic differences between missing and observed values being examined. The sensitivity analysis also explored whether or not this might have differed between randomised groups.

Because we cannot determine the true missing data mechanism based on the observed data, pattern-mixture models were implemented using MI to assess whether or not conclusions are robust to plausible departures from the MAR assumption in the sensitivity analysis. 48,49 These analyses adjusted the imputed values in the base-case analysis by either adding up to 10% to the imputed QALYs and/or total cost, or subtracting up to 10%. This sensitivity parameter was allowed to differ by group, with up to a 5% difference between the two groups (this reflects that the missing data mechanism may not be the same in the two groups, but that it is unlikely to be perfectly MAR in one group and strongly MNAR in the other). The pattern-mixture approach has been favoured in the context of clinical trial sensitivity analysis. 50,51

To explore structural uncertainty in the base-case analysis, the following scenarios around missing data for cost and QALY outcomes were explored:

1. For patients who failed to complete an EQ-5D-3L questionnaire, we assumed their HRQoL could be up to 10% lower than the MAR setting in both groups.

2. For patients who failed to complete an EQ-5D-3L questionnaire, we assumed that their costs could be up to 10% higher than the MAR setting in both groups.

3. For patients who failed to complete an EQ-5D-3L questionnaire, we assumed their HRQoL could be up to 10% lower and their costs could be up to 10% higher than the MAR setting in both groups.

4. For patients who failed to complete an EQ-5D-3L questionnaire, we assumed their HRQoL could be up to 10% lower than the MAR setting in the PDD group.

5. For patients who failed to complete an EQ-5D-3L questionnaire, we assumed their HRQoL could be up to 10% lower than the MAR setting in the WLC group.

6. For patients who failed to complete an EQ-5D-3L questionnaire, we assumed their costs could be up to 10% higher than the MAR setting in the PDD group.

7. For patients who failed to complete an EQ-5D-3L questionnaire, we assumed their costs could be up to 10% higher than the MAR setting in the WLC group.

8. A complete-case analysis was performed, in which participants with missing data were excluded from analysis.

We also explored the impact of varying the discount rate used for costs and QALYs following NICE best practice recommendations,34 ranging the discount rate from 0% to 6% per annum. Furthermore, a supplementary analysis presents costs from a wider patient/societal perspective.

Sensitivity analysis

Progression of bladder cancer

Overall, there were 19 bladder-cancer progressions in the PDD group and 12 progressions in the WL group. There were seven progressions to MIBC, 10 progressions to metastatic disease and two deaths due to bladder cancer in the PDD group. In the WL group, there were three progressions to MIBC, seven progressions to metastatic disease and two deaths due to bladder cancer. The survival curve was plotted as unadjusted Kaplan–Meier estimates (Figure 6). There was no difference in progression of bladder cancer between the PDD and WL group (HR 1.41, 95% CI 0.67 to 2.96; p = 0.369). Table 12 shows the results for the Cox proportional hazard analysis. The assumption of proportional hazards was violated (see Figure 7). The accelerated failure time model based on exponential distribution showed that there was no difference between the PDD and WL groups in terms of time to progression of bladder cancer (TR 0.64, 95% CI 0.30 to 1.36; p = 0.25).

FIGURE 6.

Kaplan–Meier survival estimates for progression of bladder cancer.

Bladder-cancer-specific survival

Overall, there were 17 deaths from bladder cancer. Of these, nine deaths were in the PDD group and eight were in the WL group. Table 13 shows the results for the bladder-cancer-specific survival using the Fine and Grey model. There was no evidence that bladder-cancer-specific survival differed between the PDD and WL groups (SHR 0.92, 95% CI 0.40 to 2.14; p = 0.852).

Overall survival

There was a total of 57 deaths, 27 in the PDD group and 30 in the WL group. Of the 57 participants who died, 17 (29.8%) died from bladder cancer, nine (15.8%) from cardiovascular events, nine (15.8%) from other cancers and 22 (38.6%) from other causes. Figure 7 shows the Kaplan–Meier survival estimates for overall survival. There was no difference in overall survival between the PDD and WL groups (HR 0.83, 95% CI 0.49 to 1.41; p = 0.496) (Table 14). The assumption of proportional hazards was violated. The accelerated failure time model based on log-normal distribution showed no evidence that the TR for trial participants differed for either the PDD or WL group (TR 1.19, 95% CI 0.70 to 2.02; p = 0.512) (see Table 13).

FIGURE 7.

Kaplan–Meier survival estimates for overall survival.

Health-related quality of life

Tables 15 and 16 report selected time points for each quality-of-life outcome; the full tables are reported with all data at all time points in Appendix 1, Tables 33 and 34.

Adjuvant therapy/intravesical treatment

Tables 17 and 18 show the adjuvant therapy received by participants post operation in those patients whose cancer recurred and in those whose cancer did not recur up to 36 months after the resection. Immediate postoperative intravesical MMC was administered to 132 (63.2%) of participants in the PDD group and 143 (65.9%) in the WL group. There was no difference between the PDD and WL groups in terms of the proportion of participants who received MMC (χ² = 0.27; p = 0.601). When compared by risk group, MMC was administered in 252 (67.4%) participants in the intermediate-risk group and 13 (40.6%) participants in the high-risk group.

In those whose cancer recurred, 17 (19.8%) participants in the PDD group and 13 (15.5%) in the WL group received BCG induction, and 13 (15.1%) participants in the PDD group and 5 (6.0%) participants in the WL group received BCG induction and maintenance. There were no differences between the PDD and WL groups in terms of the proportion of participants who received BCG induction (χ² = 0.54; p = 0.463) and proportion of participants who had BCG induction and maintenance (χ² = 3.77; p = 0.052). When compared by risk group, 39 (26.4%) participants in the intermediate-risk group and eight (50.0%) participants in the high-risk group received either BCG induction alone or BCG induction and maintenance.

In those whose cancer did not recur up to 36 months after the initial/second TURBT, one (1.5%) participant in the PDD group and four (5.4%) in the WL group received BCG induction, and 20 (30.3%) participants in the PDD group and 30 (40.5%) participants in the WL group received BCG induction and maintenance. There were no differences between the PDD and WL groups in terms of the proportion of participants who received BCG induction (χ² = 1.53; p = 0.216), and the proportion of participants who received BCG induction and maintenance (χ² = 1.59; p = 0.207). When compared by risk group, 47 (37.0%) participants in the intermediate-risk group and six (66.6%) participants in the high-risk group had either BCG induction alone or BCG induction and maintenance.

Subgroup analysis for recurrence and second resection

Of the 23 sites that participated, four (17.4%) were classified as PDD naive. Figure 8 shows the prespecified subgroup analyses comparing the outcomes from PDD-experienced and PDD-naive centres. Overall, there was no evidence that the treatment effect was moderated by PDD-naive/PDD-experienced centres. For recurrence, the HR for PDD-experienced centres was 0.91 (99% CI 0.63 to 1.31) and for PDD-naive centres it was 1.19 (99% CI 0.45 to 3.12) (interaction effect, p = 0.504). For second resection, the odds ratio for PDD-experienced centres was 0.73 (99% CI 0.37 to 1.44) and for PDD-naive centres it was 1.42 (99% CI 0.49 to 4.11) (interaction effect, p = 0.175).

FIGURE 8.

Subgroup analyses of (a) recurrence for PDD vs. WL (experienced centres vs. naive centres); and (b) second resection for PDD vs. WL (experienced centres vs. naive centres).

Bystander effect

Table 19 shows the percentage of recurrences by treatment group and surgeon experience (i.e. number of PDDs performed prior to study start). In both groups, > 50% of the participants whose surgery was performed by surgeons with experience of < 10 PDD procedures had recurrence of their bladder cancer.

Figure 9 shows the unadjusted Kaplan–Meier survival estimates by surgeon experience. Participants whose surgeries were performed by surgeons with experience of > 40 PDD procedures had a lower risk of recurrence than participants whose surgeries were performed by surgeons with experience of < 10 PDD procedures (HR 0.60, 95% CI 0.40 to 0.92; p = 0.019).

FIGURE 9.

Kaplan–Meier survival estimates for recurrence by surgeon experience.

Table 20 shows the results for the Cox proportional hazards analysis. The accelerated failure time model based on log-logistics distribution showed similar findings (TR 1.94, 95% CI 1.08 to 3.46; p = 0.026) (see Table 19).

Figure 10 shows the forest plot for the subgroup analysis by surgeon PDD experience. There was no evidence that the treatment effect was moderated by surgeon PDD experience.

FIGURE 10.

Subgroup analysis for recurrence of bladder cancer by surgeon PDD experience.

Post hoc subgroup analysis

There was no evidence that the treatment effect was moderated by EAU/EORTC risk group (Figure 11).

FIGURE 11.

Subgroup analysis for recurrence of bladder cancer by risk group at baseline.

Clavien–Dindo grade, serious adverse events and adverse events (Common Terminology Criteria for Adverse Events grade 3 or above)

The number of AEs by Clavien–Dindo grade, the number of SAEs and the number of CTCAE grade three or above events are reported in the sensitivity analysis.

There were 26 SAEs reported throughout the study (> 1 SAE could be reported per participant). There was no significant difference in the number of SAEs reported between the groups (Table 21). In total, eight participants experienced AEs (CTCAE grade 3 and above) and there was no significant difference in the number of participants who experienced an AE between the groups [rate ratio (RR) 0.62, 95% CI 0.24 to 1.60; p = 0.33]. The expected AEs following TURBT are reported in Appendix 1, Table 37.

Initial procedure

The resource use for each trial group is reported in Table 22. The total costs to the NHS, based on the microcosting approach described in Estimation of NHS costs using resource use data collected within the trial (see Table 22), are presented in Table 23. The sample size varies according to the reported data. The economic evaluation evaluates the cost-effectiveness of using PDD in the treatment of people with suspected intermediate-risk and high-risk bladder cancer. Consequently, the sample size is based on all patients randomised to either WL or PDD, whereas the statistical analysis was based on the patients with NMIBC only. There was no evidence of differences between the groups in terms of staff time and length-of-stay costs. We also conducted a sensitivity analysis that showed that a reduced unit cost (£468 instead of £891) for one night in hospital would not lead to a subsequent difference in cost (£193.62, 95% CI –£142.96 to £530.20). The additional equipment cost for PDD-TURBT is the cost of the photosensitiser (Hexvix), which results in the differences in total intervention costs between groups (£665, 95% CI £28 to £1303).

Subsequent use of services following discharge for the index procedure

Table 23 describes the use of services during follow-up. The additional costs are combined with the costs to the health services over the trial follow-up period for each treatment group; these are also presented in Table 23. Compared with WL-TURBT, PDD-TURBT does not incur additional consumable costs and additional costs are presented only. Each category of cost is presented for full cases within that category. Although PDD-TURBT is more costly than WL-TURBT treatment, there is no evidence of a difference in the total follow-up costs between the two groups.

Total NHS costs

Overall, including intervention and follow-up health-service costs, PDD-TURBT is, on average, £993 (95% CI –£724 to £2709) more costly over the 3 years’ follow-up than WL-TURBT.

Costs directly incurred by participants and indirect costs

We further incorporated both participant costs and indirect costs into the analysis over the 3-year follow-up. Table 24 reports mean costs (from a wider economic perspective) of attending inpatient admissions, outpatient appointments and primary care. Each category of cost is presented for full cases within that category. These are then summed together across all of the available cost categories for participant and companion indirect costs, and presented as the total participant cost at 3 years.

As patients with clinical symptoms of recurrence or progression attended a large number of consultations and appointments with the health services, the personal and economic costs were substantial. However, there was no evidence of differences between randomised groups, except for patient costs of accessing and using inpatient appointments. The mean differences in patient costs of accessing and using inpatient, outpatient and primary care appointments were £80.10 (95% CI £54.77 to £105.44; p < 0.001), –£37.44 (95% CI –£247.96 to £173.08; p = 0.727) and –£11.32 (95% CI –£30.34 to £7.70; p = 0.243), respectively.

Furthermore, a small proportion of patients incurred direct private health-care or self-purchased medication costs. However, the majority of patients did not and, as with the analyses above, there was no evidence of a difference between groups. The mean difference was –£10.94 (95% CI –£46.27 to £24.39; p = 0.543).

The mean indirect cost of sick leave taken by participants over 3 years for reasons related to clinical symptoms of recurrence or progression was £351.88 and £222.19 for WL-TURBT and PDD-TURBT, respectively. However, there was no evidence of a difference between the groups. The mean difference was –£129.69 (95%CI –£338.69 to £79.31; p = 0.223).

Combining all of the NHS, patient and indirect costs, we can estimate a wider overall economic cost to society. This is limited, of course, to the costs considered, and the true economic costs may be much higher. Nonetheless, the analysis gives an overall impression of the most immediate wider economic costs associated with the TURBT options considered in the PHOTO study. The total NHS, personal health-care and productivity costs were £13,193 and £14,077 for WL-TURBT and PDD-TURBT, respectively. The difference is consistent with the larger number of resections in the PDD group than in the WLC group, which results in more travel time. However, there was no evidence of a significant difference between the groups. The mean difference was £883 (95% CI –£1024 to £2788; p = 0.362).

EQ-5D-3L scores and quality-adjusted life-years

The proportion of patients with any health problems reported on the EQ-5D-3L measure of generic QoL is shown in Appendix 2, Figures 16–20. These figures present the data as reported by patients across randomised groups at baseline, discharge and follow-up visits (i.e. at 3, 6, 12, 18, 24 and 36 months), and are based on all of the available recorded data. This contrasts with the economic evaluation data in Health economic evaluation of photodynamic diagnosis of bladder tumour in reducing recurrence in primary non-muscle-invasive bladder cancer, which are based on all participants in the base-case analysis, and complete cost and QALY pairs in the sensitivity analysis. This also contrasts with the analysis of EQ-5D-3L at different time points in Chapter 4, which excludes 40 out of 135 participants in the PDD group and 49 out of 143 participants in the WLC group. The clinical analysis excluded patients who were subsequently classified as no tumour, as MIBC or who had an early cystectomy. A substantial proportion of patients appear to have had some pain or discomfort, with a significant increase after the initial surgery. Fewer patients reported problems with self-care than for the other EQ-5D-3L dimensions, with a large proportion of patients reporting no problem. A visual inspection of the graphical data does not indicate any substantial differences between the groups in any of the dimensions of generic QoL at each time point.

Table 25 provides descriptive data of mean utility scores and QALYs, generated by combining utilities with the duration (i.e. length) of life over follow-up. The preliminary utility scores for each treatment group suggest that, on average, the PDD treatment group has similar utility values over short-term (i.e. 6 months) and long-term (i.e. 3 years) follow-up. The results for incremental QALYs gained are presented, comparing PDD-TURBT with WL-TURBT for raw differences between QALY estimates. There was no evidence of a difference in QALYs gained between treatment groups at 3 years (mean difference –0.096, 95% CI –0.342 to 0.151).

Caution should be taken when interpreting Table 25, as the results are presented for only those participants who completed the EQ-5D-3L at each time point. The number of participants providing utility data in each treatment group decreased by approximately 47% from randomisation to the 1-year visit, with a further 21% decrease between the 1-year and 3-year visits.

Missing data

Missing data were mostly driven by missing EQ-5D-3L data. Utilities, as measured using the EQ-5D-3L, were completed by 89.5% and 52.1% of individuals at baseline and 36 months, respectively. Data completeness for QALYs at 3 years was evenly distributed between the groups, with data missing for 180 of 268 (67%) and 176 of 265 (66%) participants for WL-TURBT and PDD-TURBT, respectively. Furthermore, we investigated the mechanism of missingness of data by exploring the impact of baseline covariates on missing EQ-5D-3L data. Missing EQ-5D-3L data were found to differ significantly among risk categories (p < 0.01) and age groups (p < 0.01).

Complete resource use data were available for 100% of participants at the initial procedure and for 46.3%–95.7% of participants at follow-up visits of those in the ITT population. Resource use data at follow-ups were complete for 98% of participants (all of the health-care data were missing for the remaining 2% at follow-ups). Further analysis shows that the average 3-year cost of PDD-TURBT is less than that of WL-TURBT for participants without complete QALY data. This finding suggests that the complete-case analyses may overestimate the true follow-up costs for the PDD group, and that the cost difference between PDD-TURBT and WL-TURBT may be smaller after MIs.

Cost-effectiveness

Base-case analysis

The primary cost-effectiveness analysis (CEA) of the trial was conducted under the MAR assumption, using MI to impute the missing follow-up cost and HRQoL values. Effectiveness was measured in QALYs, and costs were captured by the total health-care use over the trial period. Table 26 presents the results of the base-case analysis from an NHS and PSS perspective over the 3-year time horizon. On average, PDD-TURBT is more costly and less effective; therefore, an incremental cost-effectiveness ratio (ICER) is not presented. Figure 12 illustrates the scatterplot of incremental costs and incremental QALYs for this analysis. It shows that there is substantial uncertainty in the number of QALYs gained, but also that PDD-TURBT is more costly than WL-TURBT. The CEAC in Figure 13 shows that PDD-TURBT has a 23% and 26% chance of being considered cost-effectiveness at threshold ICERs of £20,000 per QALY gained and £30,000 per QALY gained, respectively. This is a very low probability of being cost-effective. WL-TURBT is far more likely to be cost-effective than PDD-TURBT. PDD-TURBT remains dominated when using a lower unit cost (£468 instead of £891) for 1 night in hospital.

TABLE 26 - Trial-based CEA results of PDD-TURBT vs. WL-TURBT (NHS/PSS perspective)

Analysis	Adjusted, mean (95% CI)		Incremental, mean (95% CI)		ICER (£/QALY)	Probability (%) that intervention is cost-effective for different threshold values for society’s WTP for an additional QALY
	Costs (£)	QALYs	Costs (£)	QALYs		£0	£20,000	£30,000	£50,000
Base case
Imputed data analysis (3 years), MAR
WL-TURBT	12,005 (10,845 to 13,166)	2.094 (2.010 to 2.178)			WL-TURBT dominates PDD-TURBT
PDD-TURBT	12,881 (11,713 to 14,049)	2.087 (1.996 to 2.179)	876 (–766 to 2518)	–0.007 (–0.133 to 0.119)		21	23	26	30
Scenario analyses
Scenario 1: imputed data analysis (3 years), same MNAR parameters in both groups (–10% QoL)
WL-TURBT	12,005 (10,845 to 13,166)	1.956 (1.877 to 2.035)			WL-TURBT dominates PDD-TURBT
PDD-TURBT	12,881 (11,713 to 14,049)	1.948 (1.861 to 2.034)	876 (–766 to 2518)	–0.008 (–0.127 to 0.110)		21	21	24	27
Scenario 2: imputed data analysis (3 years), same MNAR parameters in both groups (+10% cost)
WL-TURBT	12,075 (10,899 to 13,251)	2.094 (2.010 to 2.178)			WL-TURBT dominates PDD-TURBT
PDD-TURBT	12,948 (11,765 to 14,132)	2.087 (1.996 to 2.179)	873 (–791 to 2538)	–0.007 (–0.133 to 0.119)		21	24	27	30
Scenario 3: imputed data analysis (3 years), same MNAR parameters in both groups (–10% QoL and +10% cost)
WL-TURBT	12,075 (10,899 to 13,251)	1.956 (1.877 to 2.035)			WL-TURBT dominates PDD-TURBT
PDD-TURBT	12,948 (11,765 to 14,132)	1.948 (1.861 to 2.034)	873 (–791 to 2538)	–0.008 (–0.127 to 0.110)		21	21	24	27
Scenario 4: imputed data analysis (3 years), different MNAR parameters in both groups (–10% QoL in PDD-TURBT group)
WL-TURBT	12,005 (10,845 to 13,166)	2.094 (2.010 to 2.178)			WL-TURBT dominates PDD-TURBT
PDD-TURBT	12,881 (11,713 to 14,049)	1.948 (1.861 to 2.035)	876 (–766 to 2518)	–0.146 (–0.269 to –0.024)		21	0	0	0
Scenario 5: imputed data analysis (3 years), different MNAR parameters in both groups (–10% QoL in WL-TURBT group)
WL-TURBT	12,005 (10,845 to 13,166)	1.956 (1.877 to 2.035)
PDD-TURBT	12,881 (11,713 to 14,049)	2.087 (1.996 to 2.179)	876 (–766 to 2518)	0.131 (0.009 to 0.254)	6664	21	85	90	93
Scenario 6: imputed data analysis (3 years), different MNAR parameters in both groups (+10% cost in PDD-TURBT group)
WL-TURBT	12,006 (10,841 to 13,171)	2.094 (2.010 to 2.178)			WL-TURBT dominates PDD-TURBT
PDD-TURBT	12,949 (11,769 to 14,128)	2.087 (1.996 to 2.179)	943 (–711 to 2597)	–0.007 (–0.133 to 0.119)		19	22	25	29
Scenario 7: imputed data analysis (3 years), different MNAR parameters in both groups (+10% cost in WL-TURBT group)
WL-TURBT	12,074 (10,903 to 13,245)	2.094 (2.010 to 2.178)			WL-TURBT dominates PDD-TURBT
PDD-TURBT	12,880 (11,709 to 14,052)	2.087 (1.996 to 2.179)	806 (–847 to 2459)	–0.007 (–0.133 to 0.119)		23	25	28	31
Scenario 8: complete-case analysis (3 years)
WL-TURBT	12,265 (10,131 to 14,399)	2.146 (2.030 to 2.261)
PDD-TURBT	15,089 (12,577 to 17,602)	2.168 (2.032 to 2.305)	3236 (–1081 to 6554)	0.034 (–0.146 to 0.213)	95,606	2	16	26	38

FIGURE 12.

Scatterplot of incremental costs and QALYs for PDD-TURBT compared with WL-TURBT: base case.

FIGURE 13.

The CEACs: base case.

Sensitivity analysis

The results under the seven missing-data scenarios (scenarios 1–7) and the complete-case analysis (scenario 8) are reported in Table 26 and as cost-effectiveness planes in Figure 14. The CEAC (Figure 15) shows that the probability of PDD-TURBT being cost-effective is relatively stable when MAR departures in total costs and HRQoL are assumed to be the same in each group (scenarios 1–3). This is also seen in Table 26, where the alternative departures from MAR had little effect on the incremental costs and QALYs in these scenarios. This will usually be the case when the missing data pattern is broadly similar across treatment groups, as the MNAR bias applies roughly equally to each group and cancels out in the treatment comparison.

FIGURE 14.

Cost-effectiveness planes under different scenarios. (a) Scenario 1; (b) scenario 2; (c) scenario 3; (d) scenario 4; (e) scenario 5; (f) scenario 6; (g) scenario 7; and (h) scenario 8.

FIGURE 15.

The CEACs under different scenarios.

Where the missing data mechanisms differ between groups (see Table 26), this suggests that the departures from MAR for the total cost (scenarios 6 and 7) would have a marginal effect on the overall results only, whereas departures for the QoL (scenarios 4 and 5) can strongly affect the conclusions. For example, PDD-TURBT appeared to be likely to be cost-effective when we assumed stronger MNAR (i.e. lower QoL) in the WL group, with a probability of being cost-effective of around 90% at £30,000 per QALY.

The results of the complete-case analysis (see Table 26) show that QALYs are similar between groups over 3 years, but PDD-TURBT is more costly. The point-estimate incremental cost per QALY gained for PDD-TURBT compared with WL-TURBT is £95,606. However, this estimate should be interpreted in the light of the considerable uncertainty surrounding it. The probability of PDD-TURBT being the preferred treatment option is substantially lower; it never reaches a probability of cost-effectiveness of > 40% at threshold values of up to £50,000 per QALY gained.

Widening the perspective of costs to include those falling on participants, families and wider societal costs changed the incremental cost to £763 (95% CI £1048 to £2574), although there were no differences between treatment groups (see Appendix 2, Table 46).

PDD-TURBT, compared with WL-TURBT, remained unlikely to be considered cost-effective over the range of society’s cost-effectiveness threshold values for a QALY that we considered; the probability that PDD-TURBT would be considered cost-effective was never above 35%.

Our results were also consistent across alternative discount rates applied to costs and QALYs (see Appendix 2, Table 48). The conclusions based on the net-benefit statistics (at a threshold value of £30,000 per QALY gained) remained unchanged for the exploration of alternative discount rates.

Changes in performance due to learning may dynamically influence the results of a technology evaluation through the change in effectiveness and costs. Learning curve analyses in the clinical analysis did not find evidence of a positive/negative learning effect. The conclusions would not change by incorporating the effect of learning in this study. Therefore, we did not conduct a learning curve analysis in the within-trial analysis.

Decision model

An economic model was produced to model the lifetime outcomes utilising the effectiveness data of the trial, but it was considered that the economic model would not provide any additional information for decision-makers. (See Appendix 3 for a description of the model.) Although the model was designed to evaluate the cost-effectiveness of PDD-TURBT, the individual patient simulation model could be edited to evaluate different surveillance strategies, accounting for the cost of a delay in diagnosis through the application of the relative risk of progression for those not receiving treatment compared with those receiving treatment during the undiagnosed period.

The economic model would be driven by the effectiveness of PDD-TURBT in terms of reducing recurrence and progression. PDD-TURBT costs more than WL-TURBT. For it to be cost-effective, PDD-TURBT needs to be more effective than WL-TURBT. The effectiveness evidence showed no statistically significant evidence of a positive effect of PDD-TURBT on recurrence or progression using either proportional hazard or accelerated failure time models. The quality-of-life and resource use outcomes were consistent with the recurrence and progression outcomes. The mean HR or TR estimates were close to 1 for recurrence and > 1 for progression. Owing to the violation of the proportional hazard assumption, the economic model would have used relative risk estimates for recurrence and progression over time periods that were different from those used in the trial, but, overall, there would be no positive effect. An economic model may estimate a greater net benefit for PDD-TURBT than the net-benefit estimate from a within-trial economic analysis if PDD-TURBT is associated with better cost and health outcomes beyond the end of the trial; however, there is no evidence for this from the PHOTO trial. As the objective of the study was to evaluate PDD-TURBT, the cost-effectiveness of different surveillance frequencies of WL-TURBT was not evaluated.

Clinical effectiveness of photodynamic diagnosis of bladder tumour in primary non-muscle-invasive bladder cancer

The PHOTO trial compared PDD-TURBT and WL-TURBT for the treatment of newly diagnosed intermediate-risk and high-risk NMIBC in a pragmatic UK setting. The main clinical outcome was bladder cancer recurrence, assessed as time to recurrence. The findings revealed no significant difference in this primary outcome measure. The CEA showed that the strategy of PDD-TURBT was not more cost-effective than WL-TURBT at 3 years. Similarly, there were no significant differences across the study’s secondary outcomes relating to cancer control, specifically cancer progression (i.e. a recurrence resulting in increased tumour staging to MIBC or metastases) and bladder-cancer-specific death. Other secondary outcomes included PROMs of HRQoL (measured using the EQ-5D-3L, EORTC-QLQ-C30 and EORTC-QLQ-NMIBC-24) and AEs (Clavien–Dindo grade, AEs and SAEs), which also showed no overall differences.

Overall, in the management of primary NMIBC, following pragmatic exclusion of those with low-risk tumours (i.e. solitary, small papillary bladder lesions of < 3 cm in diameter), the use of PDD-TURBT is not recommended.

Health economic evaluation of photodynamic diagnosis of bladder tumour in reducing recurrence in primary non-muscle-invasive bladder cancer

In the earlier evidence synthesis,11 a number of assumptions were made to link the impact of diagnosis to health outcomes. It was acknowledged in that evidence synthesis that, although PDD showed promise based on the modelling, the evidence was not conclusive given the need to splice data from multiple sources and the assumptions made about the mechanism by which costs and QALYs would be generated. It was for this reason that a trial was recommended. However, in our end-to-end evaluation, the promise seen in the modelling was not realised in practice.

Photodynamic diagnosis-guided transurethral resection of bladder tumour was, on average, a more costly procedure, driven by the cost of drugs in theatre. The drug cost excluded an associated equipment cost. It was a challenge to estimate the equipment cost because it varies depending on existing equipment and if a completely new system needs to be purchased. Including the equipment cost makes PDD-TURBT more costly. There was no evidence of a difference in the time to perform the two procedures, nor was there any evidence of a difference in terms of the follow-up care required between the groups. PDD-TURBT was more costly and less effective at 3-year follow-up from an NHS and PSS perspective than WL-TURBT. This finding did not change when different ways of handling missing data were considered, except when assuming that participants in the WL group who failed to complete an EQ-5D-3L questionnaire were likely to have been relatively poorer health.

In conclusion, over a 3-year follow-up, PDD-TURBT was, on average, more costly than WL-TURBT. There was no evidence of a difference in QALYs and it was unlikely that PDD-TURBT would be considered cost-effective compared with WL-TURBT over the range of values for society’s cost-effectiveness threshold for a QALY that we considered. These results remained unchanged over a range of plausible assumptions about missing data.

Trial limitations

Strengths of the trial

Effect of the photodynamic diagnosis experience

Those new to PDD can face technical issues related to using hardware and interpretation issues related to equivocal or false-positive fluorescence (e.g. from inflammation). Previous studies had suggested that surgeons new to PDD resection required approximately 20 cases to build experience to maximise diagnostic performance with this technology. 64,65 Accordingly, the PHOTO trial, as part of a prespecified analysis, looked to characterise and account for the potential effect of this in outcomes.

In the PHOTO trial, the effect of surgeons with previous PDD experience was compared with that of surgeons with experience of < 20 cases, who were defined as PDD naive. The outcomes for this comparison were (1) recurrence and (2) positive second resection for residual disease as a measure of complete first resection. Of 22 participating centres, four (17.4%) were classified as PDD naive. Overall, there was no evidence that the treatment effect was moderated by PDD-naive/PDD-experienced centres for either of the two outcome measures.

Evidence supporting PDD naivety affecting the ability to detect tumours when using blue-light cystoscopy is lacking; the main issue initially appears to be the detection and resection of false-positive lesions, not of missing a tumour. The phenomenon of reduced specificity was evident in a systematic review involving a total of 27 studies, enrolling 2949 participants and reporting PDD test performance. 27 In the pooled estimates for biopsy-level analysis, based on pathological confirmation, PDD had a significantly lower specificity than WLC (60%, 95% CI 49% to 71%, vs. 81%, 95% CI 73% to 90%). When surgeons start to use PDD, their surgical learning curve does not appear to affect the rate of recurrence. However, as there is no observed difference in the rate of second-resection positivity, the lack of change in recurrence rate does not seem to be related to the surgeons missing tumours because of technical naivety.

Although there appears to be convincing systematic review data (n = 27 studies; n = 2949 participants)27 using biopsy-level analysis to show that PDD has a higher sensitivity than WLC (93%, 95% CI 90% to 96%, vs. 65%, 95% CI 55% to 74%), this difference was not borne out in the PHOTO trial by reducing the rate of recurrence. An alternative explanation for this finding of no evidence of an effect may be that there is still an effect from PDD-TURBT, but that there is an associated improvement in the quality of WL resection that closes the difference in the respective event rates in each group, as explored below.

Exploration of a potential bystander effect

One possible explanation of centres reporting an initial added value of implementing PDD-TURBT, but reporting no lasting effect over time, is that practising PDD may make surgeons appreciate the shortcomings in their own ability, as well as the risk of missed tumours. This has two potential effects: by training the surgeon to (1) have a greater level of concern regarding more subtle lesions and (2) look more thoroughly, they appreciate the potential to miss lesions through experience with PDD. Ultimately, it has been speculated that this leads to a bystander effect where the quality of standard resection improves and, perhaps, accounts for observations in trials showing no difference in PDD compared with WL in highly PDD-experienced centres. 66 In the PHOTO trial, we actively planned to explore this phenomenon.

To test this hypothesis, we considered the effect on overall rates of recurrence based on surgeon’s experience of PDD. Participants whose surgeries were performed by surgeons with experience of > 40 PDD procedures had a lower risk of recurrence than participants whose surgeries were performed by surgeons with experience of < 10 PDD procedures (HR 0.60, 95% CI 0.40 to 0.92; p = 0.019). In both groups, > 50% of participants whose surgeries were performed by surgeons with experience of < 10 PDD procedures experienced a recurrence of bladder cancer. These data point to a role of PDD in improving WL-based detection and resection of NMIBC. Previously, there was a case for experience improving recurrence outcomes with TURBT, with consultants providing better outcomes than trainees. 67 The ability to improve rates of recurrence, related to PDD experience, implies that there is room to improve TURBT performance. Of note is the fact that, in sites where quality performance indicators for TURBT were implemented, no differences were observed between grades of surgeons. 68 The data from the PHOTO trial, which involved centres across the UK, provided a generalisable snapshot of TURBT outcomes and showed that improvements in TURBT are possible. These data highlight a role for better training, which could involve mandating exposure to > 40 cases of PDD-TURBT as a quality-assurance measure.

Recommendation for the disinvestment in photodynamic diagnosis-guided transurethral resection of bladder tumour

A number of previous randomised trials and systematic reviews showed increased sensitivity in the detection of NMIBC with PDD that translated into a meaningful reduction in bladder cancer recurrence. 10,11,69–73 The included trials had differing protocols, such as use of immediate postoperative intravesical chemotherapy, second resections or approaches to adjuvant intravesical treatments (i.e. chemotherapy and BCG), making it difficult to extrapolate these findings to current practice. 11 Furthermore, differences in inclusion criteria, trial design and outcome measurement limit accurate meta-analytical comparison. 11 Nevertheless, these previous data had led to the uptake of PDD technology across the UK, showing added value in reducing recurrence in ‘real-life’ practice when comparing cohorts treated with PDD and those treated with high-quality WL-TURBT. 73,74 Similarly, there was uptake across Europe and the USA, where strong expert recommendations were made on both the reduction of recurrence and health economic evaluations. 75–78 However, to further substantiate the recommendations from efficacy trials and the subsequent non-randomised reporting of ‘real-life’ practice, a pragmatic randomised trial was required, taking into account current standard treatment approaches, including immediate postoperative intravesical chemotherapy, second resections or adjuvant intravesical treatments.

The PHOTO study addresses the outstanding questions outlined above, providing clear and strong evidence that, in primary intermediate- and high-risk NMIBC managed in the day-to-day UK setting, there is no reduction in recurrence at 3 years with PDD resection compared with standard-of-care WL resection. The main aim of this trial was to provide a precise, unbiased measure of benefit for PDD-TURBT in reducing recurrence and it was rigorously designed accordingly. The highest quality evidence regarding clinical effectiveness is essential, as systematic reviews informing clinical guidance recommendations are affected by the risk of bias and low methodological quality of previous trials. The PHOTO trial findings do not support the use of PDD resection for primary intermediate- and high-risk NMIBC.

Up-to-date information for clinicians, patients and service providers on current treatments for non-muscle-invasive bladder cancer, their outcomes and costs

Although guideline recommendations are designed to use evidence-based medicine to instruct best clinical practice, their implementation can be patchy. Some of the reasons for this are that there can be limitations in the quality of the information available to inform guidelines or that the recommendations are not acceptable/relatable to current clinical practice. 79,80 For bladder cancer guidelines, such as those of NICE,18 EAU12 and the American Urological Association (AUA),59 we have an excellent resource, comprising authoritative and comprehensive systematic reviews, meaning that advice is generally consistent across these major guidelines and is well accepted. 81,82 Despite the outstanding standard of these guidelines, questions remain about the expected outcomes in treating NMIBC and they are essential to counselling patients regarding options and outcomes. This is mainly because there is a strong reliance on historic EORTC data to inform NMIBC management, despite recent efforts to update the prognostic risk groups. 61 Here, we can now provide new up-to-date data based on contemporary management and outcomes of NMIBC in accordance with current NICE18 and EAU guidance. 12 These new data are further explored below.

Other clinical utility for photodynamic diagnosis in non-muscle-invasive bladder cancer

Despite our findings in the specific context of the trial question, the role for PDD-TURBT in other contexts remains an open question. We have shown that PDD-TURBT is safe and well tolerated, with rates of complication and AEs that are no different from conventional WL-TURBT. The current EAU guideline12 recommends PDD in certain clinical situations, graded as strong recommendations, which remain areas for future research to generate more substantial evidence to support their role. The EAU guideline describes taking biopsies from visually abnormal regions and normal-looking epithelium when urine cytology is positive, non-papillary tumours are seen or there is a previous history of high-grade cancers. The guideline states that, if available, PDD-guided biopsies should be taken. In cases where mapping biopsies are indicated, such as those with positive cytology and normal-looking cystoscopy with no upper-tract urothelial cancer, the EAU recommends that PDD-guided biopsies can be used instead of mapping biopsies. Similarly, in the follow-up of patients with normal cytoscopy but positive cytology, a PDD-guided biopsy, where available, can be used instead of a mapping biopsy.

The next generation of cystoscopic bladder imaging with or without novel photosensitisers

The accurate detection of bladder tumours remains a major clinical need and an active research focus. Future research regarding light technologies (e.g. narrow-band imaging, multispectral imaging, optical coherence tomography and artificial intelligence-enhanced visualisation) and the next generation of photosensitiser technology remain of interest to improve the detection and staging of NMIBC. 91–95

Rapid biomarker assessment

The PHOTO trial’s health economic model contains up-to-date UK costs associated with the NMIBC management pathway. This model establishes diagnostic and cost thresholds that could be presented to make a case for NICE guideline implementation and could inform rapid assessment of new biomarkers to replace cystoscopy for diagnosis or more accurate and/or effective approaches to resection (e.g. adjuvant treatments such as immuno-oncology, or novel imaging or resection approaches such as en bloc laser resection).

There are a number of established, commercially available urinary biomarkers [e.g. BTA STAT^® (Polymedco Inc., Cortlandt, NY, USA), BTA TRAK^® (Polymedco Inc.), Alere NMP22 BladderCheck^® (Abbott Molecular, Abbott Park, IL, USA), ImmunoCyt/UCyt+™ test (Diagnocure Inc., Quebec City, Canada) and UroVysion Bladder Cancer Kit™ (Abbott Molecular)] and a plethora of emerging next-generation ‘omic’-based tests that include urinary deoxyribonucleic acid (DNA) and proteomic reads. 96 Rapid urinary biomarker assessment for the presence of bladder cancer followed by cystoscopy and resection for positive biomarker cases could be a cheaper, but possibly less accurate, method for diagnosing NMIBC. Therefore, there is uncertainty in the cost-effectiveness of this diagnostic approach. False-negative results from rapid biomarker tests could result in a delay in diagnosis until the cancer had progressed, as the cancer would not be identified until either an opportunistic retest or the patient became symptomatic. The cost-effectiveness of a rapid biomarker test in this role would depend on the proportion of cystoscopies that could be avoided by using the biomarker test. The PHOTO trial’s health economics model could, potentially, be adapted to evaluate the cost-effectiveness of a rapid biomarker test in this role and the budget impact of a cheaper diagnostic approach. Evidence of the diagnostic accuracy of a rapid biomarker test in this role compared with cystoscopy would be required. Evidence on the stage of bladder cancer for symptomatic and asymptomatic evidence would also be required. If the cancer stage remains non-invasive when the patient becomes symptomatic, then evidence on the time to recurrence for symptomatic compared with asymptomatic NMIBC patients would also be useful to characterise. For any rapid biomarker test to be cost-effective, the risk of a missed diagnosis and the delay to diagnosis would need to be low. Some of these data for specific tests are emerging in the literature and would be of interest to explore in the PHOTO trial’s health economic model in future work.

Modelling alternative surveillance strategies

The cost-effectiveness of different cystoscopic surveillance schedules, possibly in combination with the rapid biomarker tests discussed in the preceding section, could also be evaluated as alternatives to the current surveillance schedule approach by adapting the PHOTO trial’s health economics model. The current schedule for cystoscopy follow-up is based on expert recommendation, relating to the observation that most recurrences occur early, which is apparent in the sharp drop-off shape of the Kaplan–Meier curves for recurrence-free survival presented in this report. There are now emerging data looking at conditional probabilities for recurrence, where the longer an individual goes without recurrence, the more likely they are to avoid recurrence. 97 These data, along with our contemporary recurrence-free survival and those of other up-to-date studies, such as BOXIT, could lay the foundation of a ‘lighter’, dynamic, risk-adaptive follow-up protocol that could produce savings in the massive finance burden currently incurred with cystoscopy surveillance.

In future research, the rapid biomarker test could be evaluated to replace cystoscopy either at every time point in the surveillance schedule or at some time points (e.g. every other time point). Once again, the rapid biomarker test surveillance strategy could be cheaper than the cystoscopy surveillance strategy, but the delayed diagnosis in some cases would result in worse health outcomes for some patients. For the rapid biomarker test to be cost-effective, the risk of a missed diagnosis would need to be low and the delay to diagnosis would need to be small. This analysis would require similar evidence to that described above, although it is possible that the diagnostic accuracy of the rapid biomarker test may differ during surveillance from the accuracy at initial diagnosis.

Updated non-muscle-invasive bladder cancer recurrence, progression and survival prediction tools

As discussed above, the tools currently available for predicting clinical outcomes are reliant on historical data and appear to overestimate recurrence, progression and survival, as shown in our PHOTO trial data. Previous assessments of EORTC prognostic groups performance have been tested in large, external, Spanish and US data sets. 63,98 Data on 1062 patients with NMIBC treated with BCG in the CUETO clinical trials were analysed, showing that the EORTC groups successfully stratified recurrence and progression risks, but overestimated risks of recurrence and progression after BCG therapy. 63 Interestingly, the CUETO scoring system exhibited similarly poor discrimination for recurrence and progression. 99 These overestimations remained in BCG-treated patients, especially for the EORTC tables. External validation of the EORTC risk calculator, in comparison to the National Comprehensive Cancer Network (NCCN) risk groups, was also undertaken in a contemporary US population of 1491 NMIBC patients; it showed that there is overestimated progression among the highest-risk group, which is consistent with prior publications. 98 Additional studies of the EORTC tables’ performance have included a comparison with the simplified, treatment-directed, EAU risk group stratification (which are also based on the EORTC tables) in a multi-institutional database of 5122 patients. The EAU categories reclassified 37.9% of patients into a higher risk group for recurrence and 11.8% of patients into a higher risk group for progression, but, pragmatically, assigned most patients to the same treatment recommendations.

The main issues identified with these established tools were that none used data that reflected current standards of treatment, and recommendations were made to update scoring models with previously unavailable data. 58 In response, updated EAU prognostic factor risk groups were defined from 3401 retrospectively collected individual patient’s data for primary NMIBC patients from the institutions of members of the guidelines panel. 61 Multivariable Cox proportional-hazards regression models were fitted to the primary end point: the time to progression to muscle-invasive disease or distant metastases. A new, very high-risk group was identified and lower risks were assigned for progression to the remaining low, intermediate and high groups. Herein lies an opportunity to combine data from contemporary, prospective, randomised trials (e.g. PHOTO, BOXIT53 and more recent EORTC trials addressing BCG treatment in higher risk individuals100) to generate a more robust prediction tool. Models including competing risk of death would also provide utility when making clinical judgements regarding treatments with a high morbidity rate, such as early cystectomy.

Translational projects accessing the tissue biobank (PHOTO-T project)

Previously, experts have reported accurate visual diagnosis of cancerous compared with benign tissue, and reliable prediction of the stage and grade of cancer; however, in our ‘real-world’ pragmatic study, visual diagnosis was not shown to be reliable using cystoscopy alone. New, more accurate approaches involving biomarkers could be developed for this assessment and be based on urine containing cancer cells that have shed from the tumour or cellular contents [i.e. DNA, ribonucleic acid (RNA) or protein]. There is already a growing body of evidence providing diagnostic tools and the ability to distinguish NMIBC from MIBC. 96 In addition, these tools could have a role in earlier diagnosis, staging and expedited management of bladder cancer. To accelerate the development of such biomarkers, well-annotated tissue biobanks with detailed clinic-pathological data are required; this formed the rationale for archiving urine, blood, and bladder tissue from consenting PHOTO trial participants as part of the PHOTO-T project.

There are a number of separately funded translational projects currently accessing the PHOTO-T archive, either (1) developing new biomarkers {e.g. using tumour-specific mitochondrial deoxyribonucleic acid (mtDNA) mutations (Biomedical Research Centre funded) and exploring urinary volatile agents for diagnosis [Cancer Research UK (CRUK) funded]} or (2) providing external validation of diagnostic markers [e.g. Medical Research Council (MRC)- and CRUK-funded projects looking at urinary DNA methylation and RNA markers]. In the future, other projects may look to access this resource for new discovery or validation work.

Enhanced training in cystoscopy

In the PHOTO trial, we saw an interesting phenomenon in which the use of PDD in previously PDD-naive centres saw a reduction of cancer recurrence in their WL-treated participants. This bystander effect of PDD improving WL detection of tumour and its resection requires further exploration. If this effect is confirmed, PDD may provide a tool for enhancing the clinical training of junior urologists, producing a better standard of WL resection for the next generation of urologists and patents.

Long-term outcome data collection

The PHOTO study has detailed clinical outcomes and costs based on 3 years’ follow-up, and extended those analyses to include a modelled lifetime projection for the health economic assessments. Ideally, we would like to collect longer-term outcomes to more accurately inform our lifetime model. Initially, we had proposed linkage of the trial data with the national British Association of Urological Surgeons (BAUS) Oncology Data and Audit registry101 (Nuvola) over the next 10 years to further inform the HTA model; however, this registry is now being dissolved. There are ongoing discussions within BAUS to establish mechanisms for future procedure and outcomes data. Possible ‘snapshot’ audits, in which centres look at longer-term outcomes for patients and that could include a focus on those enrolled in the PHOTO trial, have been proposed by the BAUS Audit Steering Group. It is most likely that separate grant applications will look to update the HTA model with longer-term data in time, aiming to link to Public Health England (PHE)-maintained National Cancer Registration and Analysis Service (NCRAS). These longer-term data would further support the PHOTO trial’s health economic model in making more accurate 5–10 year predictions for changes to improve the NMIBC pathway.

Recruitment sites

We would like to thank the staff and participants at the following centres:

• NHS Grampian (Aberdeen, UK) – Sarfraz Ahmad (2015–19) and Justine Royle (PI)

• Royal Free London NHS Foundation Trust (London, UK) – Paras Singh (PI)

• Hampshire Hospitals NHS Foundation Trust (Basingstoke, UK) – Hugh Mostafid (PI)

• Hull University Teaching Hospitals NHS Trust (Hull, UK) – Mathew Simms (PI)

• Imperial College Healthcare NHS Trust (London, UK) – Norma Gibbbons (PI)

• Oxford University Hospitals NHS Foundation Trust (Oxford, UK) – Jeremy Crewe (PI)

• Dartford and Gravesham NHS Trust (Dartford, UK) – Sanjeev Madaan (PI)

• Newcastle Hospitals NHS Foundation Trust (Newcastle upon Tyne, UK) – Andrew Thorpe (PI)

• South Tees Hospitals NHS Foundation Trust (Middlesbrough, UK) – Joanne Cresswell (PI)

• East and North Hertfordshire NHS Trust (Stevenage, UK) – Nikhil Vasdev (PI)

• Swansea Bay University Health Board (Swansea, UK) – Jonathan Featherstone (2015–16) and Pradeep Bose (PI)

• NHS Tayside (Dundee, UK) – Ghulam Nabi (PI)

• Cwm Taf Morgannwg University Health Board (Bridgend, UK) – Rhidian Hurle (PI)

• Derby Teaching Hospitals NHS Foundation Trust (Derby, UK) – Amjad Peracha (PI)

• Royal Devon and Exeter NHS Foundation Trust (Wonford, UK) – John McGrath (PI)

• University Hospitals of North Midlands NHS Trust (Stoke-on-Trent, UK) – Lyndon Gommersall (PI)

• Salisbury NHS Foundation Trust (Salisbury, UK) – Melissa Davies (PI)

• University Hospital Southampton NHS Foundation Trust (Southampton, UK) – Bhaskar Somani (PI)

• Leeds Teaching Hospitals NHS Trust (Leeds, UK) – Sunjay Jain (PI)

• Ashford and St Peter’s Hospitals NHS Foundation Trust (Ashford, UK) – Sachin Agrawal (PI)

• University College London Hospitals NHS Foundation Trust (London, UK) – Mark Feneley (PI)

• NHS Lothian (Edinburgh, UK) – Param Mariappan (PI).