Systematic review of the clinical effectiveness and cost-effectiveness of photodynamic diagnosis and urine biomarkers (FISH, ImmunoCyt, NMP22) and cytology for the detection and follow-up of bladder cancer

Introduction

Bladder cancer, or more precisely malignant neoplasm of the bladder,1 is a disease in which the cells lining the urinary bladder lose the ability to regulate their growth and start dividing uncontrollably. 2 This abnormal growth results in a mass of cells that form a tumour. People with a suspicion of bladder cancer mainly present with urinary symptoms including gross haematuria, microscopic haematuria and urinary tract symptoms. Bladder cancers can be broadly categorised into two main groups depending upon their extent of penetration into the bladder wall: non-muscle invasive and muscle invasive. The majority of diagnosed patients (75–85%) present with non-muscle-invasive disease, which as described in the next subsection is characterised by a probability of recurrence at 5 years from 31% (95% CI 24% to 37%) to 78% (95% CI 73% to 84%) despite treatment. 3 The remaining cancers are muscle invasive and/or metastatic.

Aetiology, pathology and prognosis

Epidemiology

Bladder cancer is the sixth most common cancer in the UK. 22 Bladder cancer is the most frequently occurring tumour of the urinary system and accounts for 1 in every 28 new cases of cancer diagnosed each year in the UK. During the last three decades there has been a gradual decrease in the incidence of bladder cancer (Figure 1). 22 However, changing trends in the incidence of bladder cancer over time are difficult to interpret because of different and changing classifications and coding practices of the condition. 5

FIGURE 1.

Age-standardised (European) incidence rates of bladder cancer by sex, UK, 1993–2004.

Incidence and prevalence

Bladder cancer is the fourth most common cancer in men and the tenth most common in women in the UK. 22 In 2005, the estimated male and female crude incidence rates of bladder cancer were 24.6 and 9.3 per 100,000 population with 6091 and 2403 new cases, respectively, in England, and 43.0 and 17.2 per 100,000 population with 619 and 260 new cases, respectively, in Wales (Table 2). 22

Although the overall incidence of bladder cancer in the UK has remained much higher in men than in women in the last five decades, it has shown a slow decrease between 1993 and 2005 (Figure 1) following a rapid rise between 1971 and 1993. 22,23 In addition, in England and Wales, the prevalence of bladder cancer increased by 57% between 1971 and 1998, particularly in women. 23

Variation in incidence by age

The mean age at which bladder cancer is diagnosed in the UK is 71.3 years. The incidence and mortality rate of bladder cancer rapidly increase with increasing age (Figures 2 and 3). Bladder cancer commonly occurs in older people and is rare in people under 50 years of age.

FIGURE 2.

Numbers of new cases and age-specific incidence rates of bladder cancer by sex, England and Wales, 2005.

FIGURE 3.

Numbers of deaths from and age-specific mortality rates of bladder cancer by sex, UK, 2006.

Impact of the health problem

Diagnosis

Haematuria is presence of blood in the urine and is the most common symptom of bladder cancer. Bladder cancer is detected in approximately 10% of patients with gross haematuria and 3–5% of those with microscopic haematuria aged over 40 years. 25,26 Less commonly, individuals may note disturbance in their urinary habits including complaints of dysuria (painful urination), increased frequency, urgency of urination, failed attempts to urinate and urinary tract infection. These symptoms can raise suspicion of diffuse CIS. Other symptoms that may be attributed to a mass in the bladder or ureteral obstruction are likely to indicate that bladder cancer may be muscle-invasive disease. 5,24,27

Management of disease

The management of non-muscle-invasive bladder cancer is based on: (1) the pathological findings of the biopsy specimen, with attention to histological type, grade and depth of invasion; (2) the presence of associated CIS; (3) the number of tumours; (4) previous recurrence rate if applicable; and (5) size of tumour. Depending on these findings, treatment options include cystoscopic follow-up only (either flexible or rigid cystoscopy under general anaesthesia), cystoscopic follow-up and intravesical chemotherapy and immunotherapy courses or radical cystectomy.

The goals of current treatment for patients with non-muscle-invasive bladder cancer are to prevent disease recurrence or progression to muscle-invasive disease to avoid loss of the bladder and, ultimately, to enhance survival. The current treatment strategies for patients with bladder cancer depend on three main types of bladder cancer, non-muscle-invasive disease, muscle-invasive disease and metastases, as recommended in the multidisciplinary team (MDT) guideline. 28

Current service cost

It is difficult to estimate the current bladder cancer service cost in the UK because of the variation in practice in the diagnosis and follow-up of patients based on their risk categorisation. It is anticipated that the costs of the higher risk patients will be greater than those of the low-risk patients because of more follow-up interventions. The total cost of treatment and 5-year follow-up of patients with bladder cancer diagnosed during 2001–2 was £55.39 million; the total cost of superficial disease was £35.25 million and that of invasive disease was £20.2 million. The total cost for patients undergoing radical radiotherapy was over twice that for those undergoing cystectomy (£8.1 versus £3.6 million)32 In the USA it is estimated that $1.7 billion is spent on bladder cancer. 33

An estimate of the current cost to the UK NHS can be generated by using the total cost of each strategy (see Tables 39 and 42) and combining it with the values in Table 2. If it assumed that the current practice for diagnosis in the UK is flexible cystoscopy and cytology for initial diagnosis followed by white light rigid cystoscopy [CSC_CTL_WLC (CSC_WLC)] the cost per low-risk patient will be £6302.25. Therefore the total annual cost to the NHS will be £64,765,481. There is also evidence that costs are likely to increase with improved survival because patients need several courses of treatment.

Variation in services and/or uncertainty about best practice

All urology departments offer haematuria clinics and subsequent TURBT if appropriate either in the same hospital or in a hub hospital. Radiotherapy and systemic chemotherapy are available in cancer centres. Radical surgery for prostate and bladder cancer should be provided by teams carrying out a cumulative total of at least 50 such operations per annum. These procedures should be performed by surgeons performing at least five of either radical cystectomy or prostatectomy each year. 34

Relevant national guidelines, including National Service Frameworks

The relevant national guidelines are:

• National Institute for Clinical Excellence (2002). Improving outcomes in urological cancers. NHS guidance on cancer services34

• National Institute for Clinical Excellence (2003). Laparoscopic cystectomy of the urinary bladder. IPG02635

• Scottish Intercollegiate Guidelines Network (SIGN) (2005). Management of transitional cell carcinoma of the bladder36

• National Institute for Health and Clinical Excellence (2007). Intravesical microwave hyperthermia with intravesical chemotherapy for superficial bladder cancer. IPG23537

• NHS Pan-Birmingham Cancer Network (2006/2007). Guidelines for the management of bladder cancer38

• UK National Screening Committee (NSC) (2002). Evaluation of urinary tract malignancy (bladder cancer) screening against NSC criteria39

• British Association of Urological Surgeons (BAUS) Section of Oncology and Uro-oncology Group (2007). MDT (multi-disciplinary team) guidance for managing bladder cancer28

• European Association of Urology (EAU) (2009). Guidelines on TaT1 (non-muscle-invasive) bladder cancer3

• European Association of Urology (EAU) (2009). Guidelines on bladder cancer: muscle invasive and metastatic40

• American National Comprehensive Cancer Network (NCCN) (2009). NCCN clinical practice guidelines in oncology. Bladder cancer including upper tract tumours and urothelial carcinoma of the prostate24

• American Urological Association (AUA) (2007). Guideline for the management of nonmuscle invasive bladder cancer (stages Ta,T1 and Tis). 5

Only two of the above guidelines specifically mention photodynamic diagnosis (PDD):

The evidence suggests potential benefits from photodynamic techniques for patients with superficial bladder cancer undergoing initial resection of their tumour. Its role in patients developing recurrence during followup is less clear.

SIGN (2005)36

The benefit of fluorescence-guided TURBT for recurrence-free survival was shown in several small randomised clinical trials, but its value remains to be proven in improving the outcome of patients for progression rates or survival. The additional costs of the equipment should be considered.

EAU (2009)3

Various guidelines, including those of the EAU and AUA, recommend the use of voided urinary cytology, both in the diagnosis and surveillance of non-muscle-invasive bladder carcinoma. However, there are no equivalent recommendations for the use of biomarkers. Although the international consensus panel on the use of biomarkers in bladder cancer realised the importance of non-invasive diagnosis and surveillance of non-muscle-invasive disease, it concluded that, although none of the non-invasive tests could replace cystoscopy, many markers together with cystoscopy could improve the current practice of managing patients with bladder cancer. 41

Summary of interventions

Identification of important subgroups

Current usage in the NHS

Anticipated costs associated with the technologies

The anticipated costs associated with the technologies will depend on the strategies used in the diagnosis and follow-up of patients. The average unit cost of diagnosing bladder cancer using PDD is £1371, rigid white light cystoscopy £937, flexible cystoscopy £441, cytology £92.37, NMP22 £39.3, FISH £54.8 and ImmunoCyt £54.8; and the cost of treatment using PDD-assisted TURBT is £2436, WLC-assisted TURBT £2002, mitomycin £73, BCG £89, cystectomy £6856, chemotherapy £50.22, radical radiotherapy £1050 and palliative treatment £12,825 (see Chapter 6 for details). The modelling results indicate that using the most effective strategy (the one with the highest number of true positives and the lowest number of false negatives), which includes either of the two biomarkers FISH or ImmunoCyt and PDD as the initial strategy and either FISH or ImmunoCyt with WLC as the follow-up strategy, will cost £5919.28 per low-risk patient per year.

Inclusion criteria (see Chapter 3)

Key issues

The key issues to be addressed are:

• Can PDD improve detection of bladder cancer (1) at the time of TURBT for newly diagnosed disease and (2) during follow-up of patients with non-muscle-invasive disease?

• Can PDD reduce recurrence and/or progression of non-muscle-invasive bladder cancer compared with WLC?

• Can urine biomarkers (FISH, ImmunoCyt, NMP22) improve detection of bladder cancer during (1) initial diagnosis of patients suspected of having bladder cancer and (2) follow-up of patients diagnosed with non-muscle-invasive disease?

• What is the incremental cost-effectiveness of PDD during TURBT for newly diagnosed non-muscle-invasive bladder cancer and during follow-up?

• What is the incremental cost-effectiveness of biomarkers during the initial diagnosis of patients suspected of having bladder cancer and during follow-up of those diagnosed with non-muscle-invasive disease?

Care pathways

Care pathways describing plausible strategies for the initial diagnosis and follow-up of people with bladder cancer were developed. The basic care pathway was based on discussions with the clinical experts involved in this study and a brief description of this is provided within Chapter 1.

Initial diagnosis and treatment (Figure 4)

The pathway begins with an initial presentation of symptoms or asymptomatic microscopic haematuria and varies in terms of where and when biomarkers and PDD might be used. Patients who present with either microscopic or gross haematuria or lower urinary tract symptoms are tested using flexible cystoscopy and cytology. Biomarkers could be used at this point either in addition to these two tests or instead of cytology. The results of these tests can be either negative or positive. Patients who have two/three negative results are discharged. Discharged patients who later re-present with similar symptoms go back to the beginning of the care pathway. Patients with one or more positive results for these tests as outlined in Table 3 undergo TURBT during which PDD may be used with the aim of improving the detection of tumours, thereby potentially reducing the rate of residual tumours and increasing the detection of CIS and small papillary tumours.

TABLE 3 - Different test results

–, negative; +, positive.

Cystoscopy	Cytology	Biomarkers
–	–	–
–	–	+
–	+	–
+	–	–
–	+	+
+	–	+
+	+	–
+	+	+

After TURBT is performed for newly diagnosed bladder cancer, the standard UK management is that the patient also receives a single instillation of adjuvant intravesical mitomycin C, ideally within 6 hours of resection but not later than 24 hours if possible. Biopsies are taken and the results of the histological analysis may be either negative or positive for bladder cancer. Those who have a negative histology result are then discharged. Discharged patients whose symptoms are not resolved may subsequently re-present at the beginning of the care pathway. For the purposes of this review, although patients who have a negative bladder cancer test result are considered as discharged, it is noted that some who initially had a positive result may be at risk of upper tract urothelial cancer or renal cancer and consequently will require further tests, and, if positive, treatment.

Those patients whose histological results confirm the presence of bladder cancer are classified into muscle-invasive or non-muscle-invasive disease. For those with muscle-invasive disease, treatment options are outlined in Figure 4. Essentially, those amenable to potential cure are offered either radical cystectomy with bilateral pelvic lymphadenectomy or radiotherapy. Treatment with surgery or radiotherapy is usually preceded by three cycles of systemic neoadjuvant cisplatin-based chemotherapy. The rationale for chemotherapy is that over 50% of patients with muscle-invasive disease have occult metastatic disease at presentation. It is noted that practice at individual centres may vary. The decision for cystectomy or radiotherapy is primarily based on patient choice and medical fitness. The presence of concomitant CIS and upper tract dilatation are also factors that favour cystectomy. For patients with more advanced metastatic disease, the treatment is palliative.

FIGURE 4.

Developed care pathway – initial diagnosis/treatment. BM, biomarkers; CSC, flexible cystoscopy; CTL, cytology.

Follow-up of patients with non-muscle-invasive bladder cancer (Figure 5)

The key factors increasing the risk of recurrence and progression in patients with non-muscle-invasive bladder cancer are: (1) tumour multiplicity, (2) greater tumour diameter, (3) previous recurrence rate, (4) higher T-stage, (5) concomitant CIS and (6) higher histological grade. A brief summary is provided in the following sections and a further short review on the management of bladder cancer, required for the description of the model structure, is provided in Chapter 6 (see Model structure, Markov model).

High risk

Broadly speaking, patients with Ta/T1G3 TCC, CIS or multifocal T1G2 TCC are classified as being at high risk of not only recurrence but also progression. If diagnosed with T1G3 TCC they are offered an early re-resection to ensure that they are not muscle invasive. All patients in this group are usually offered an induction course of six intravesical BCG instillations followed by a maintenance regimen of a further 21 instillations over a 3-year period. Some may opt for primary radical cystectomy. Patients who opt for bladder sparing undergo their first bladder check at 3 months. If they remain tumour free they are followed up every 3 months for the first 2 years and then every 6 months thereafter. During the follow-up visits, patients undergo cystoscopy and in some centres cytology and/or a biomarker test. Patients found to have a non-muscle-invasive recurrence have four options: they can undergo cystectomy, have a second induction course of BCG and then reassess, have three further instillations of BCG and then reassess, or receive endoscopic control.

Low and intermediate risk

Patients at low risk of recurrence and progression have TaG1 TCC or solitary T1G1 TCC. Those at intermediate risk have TaG2 TCC or multifocal T1G1 TCC. Multiplicity at presentation and a tumour recurrence at 3 months have consistently been shown to be key practical predictors of future recurrence, and many urologists in the UK tailor their cystoscopic follow-up of low- and intermediate-risk patients based on these two factors for these reasons:

1. Patients who have a solitary tumour at diagnosis and no tumour recurrence at 3 months are followed up at 9 months and then annually for 4 further years. If at the end of this 5-year follow-up period they have remained tumour free they are discharged. During the follow-up visits these patients undergo flexible cystoscopy and in some centres cytology and/or biomarker tests. Although most patients with a tumour recurrence will receive TURBT, some may have a cystodiathermy and biopsy.

2. Patients with multiple tumours at presentation and no recurrence at 3 months or a solitary tumour at presentation with recurrence at 3 months need more intense follow-up and are followed up every 3 months for the first year and annually if they remain tumour free until 10 years and are then discharged. During the follow-up visits patients undergo cystoscopy and in some centres cytology and/or biomarker tests. Those who present with a tumour at the follow-up visit undergo either TURBT or cystodiathermy and biopsy. These patients may be considered for a course of six intravesical instillations of mitomycin C or epirubicin.

3. Patients with multiple tumours at presentation and recurrence at 3 months have the highest risk of recurrence and are followed up every 3 months for the first 2 years and then annually thereafter. They are usually offered a course of six intravesical instillations of mitomycin C or epirubicin. Those who present with a tumour at the follow-up visit undergo either TURBT or cystodiathermy and biopsy. During the follow-up visits patients undergo cystoscopy and in some centres cytology and/or biomarker tests. Cystoscopies in the first 2 years are usually under general anaesthesia using a rigid cystoscope.

During the follow-up period the status of patients may change and they may develop muscle-invasive tumours. It is also possible that patients may die at any time during follow-up from causes related to bladder cancer or from unrelated causes. The outlined care pathways in Figures 4 and 5 identify the areas in which PDD and biomarkers could be used in conjunction with the standard tests to diagnose patients with suspected bladder cancer and to follow up those who have been diagnosed with non-muscle-invasive disease. These patient care pathways will be used to inform the economic model and to establish whether the use of PDD and urine biomarkers reduces recurrence or decreases progression at follow-up as a consequence of altered treatment.

FIGURE 5.

Developed care pathway – follow-up. BM, biomarkers; CSC, flexible cystoscopy; CTL, cytology.

Types of studies

The types of studies considered for reporting test performance were:

• direct (head-to-head) studies in which the index test and reference standard test were performed independently in the same group of people

• randomised controlled trials (RCTs) in which people were randomised to the index and comparator test(s) and all received the reference standard test.

In the event that there was insufficient evidence from direct and randomised studies we considered undertaking indirect (between-study) comparisons by meta-analysing studies that compared each single test or combination of tests with the reference standard test, and making comparisons between meta-analyses of the different tests. However, this type of study design is less reliable than direct studies as differences in diagnostic accuracy are susceptible to confounding factors between studies. The following types of studies were considered:

• Observational studies, including case series, in which the sample is created by identifying all people presenting at the point of testing (without any reference to the test results).

• Case–control studies in which two groups are created, one known to have the target disease and one known not to have the target disease, when it is reasonable for all included to go through the tests. We excluded case–control studies when the control group consisted of completely healthy volunteers, or when the control group consisted of completely healthy volunteers and people with benign urinary conditions and it was not possible to calculate results for the control group minus the healthy volunteers, such that the spectrum of disease and non-disease was unlike that to be encountered in a diagnostic situation.

Studies reporting test performance had to report the absolute numbers of true positives, false positives, false negatives and true negatives, or provide information allowing their calculation. Studies reporting patient- and/or biopsy-level analysis (for PDD) and patient- or specimen-level analysis (for biomarkers/cytology) were considered.

For assessment of the effectiveness of PDD-assisted TURBT compared with WLC-assisted TURBT in terms of outcomes such as recurrence or progression we focused on RCTs.

Types of participants

The participants considered were people (1) suspected of having bladder cancer or (2) previously diagnosed with non-muscle-invasive bladder cancer and having follow-up cystoscopic examination.

Index and comparator tests

The following tests and comparators were considered:

• PDD (using the photosensitising agents 5-ALA, HAL or hypericin) compared with WLC

• urine biomarkers (FISH, ImmunoCyt, NMP22) or cytology either alone or compared with each other.

Studies reporting the test performance of combinations of the above tests were also considered.

If the evidence allowed, the following subgroup analyses were planned:

• number of tumours on first cystoscopic examination

• type (e.g. CIS) and grade of tumour (WHO 1973 or 2004 classification)

• tumour recurrence at the first 3-month cystoscopic examination following TURBT

• diagnostic performance of the different PDD photosensitising agents

• diagnostic performance of the different categories of urine biomarkers

• for urine biomarkers, whether the urine sample was voided or obtained by bladder wash.

Numerous biomarkers exist that potentially could have been included in the review but to make the task manageable within the given time frame the review’s steering committee agreed that the review should focus only on those biomarkers regarded as being most clinically relevant. These were seen as being either those approved by the US FDA or the three generally regarded as most useful – FISH, ImmunoCyt and NMP22 – with cytology also included. It was agreed that the Chairman of the BAUS Section of Oncology should be contacted to canvass the views of the Section’s Executive Committee on the most relevant biomarkers to consider. Following this, the Chairman on behalf of the Section suggested that the review should assess ImmunoCyt, NMP22, FISH and cytology, and consequently these were the tests that were included in the review.

Reference standard

The reference standard considered both for studies reporting PDD and for studies reporting biomarkers was histopathological examination of biopsied tissue.

Types of outcomes

The following outcomes were considered:

• for PDD:

  1. – test performance in detecting non-muscle-invasive bladder cancer

  2. – recurrence rate of bladder tumour over time following initial resection

  3. – progression to muscle-invasive disease

• for urine biomarkers/cytology:

  1. – test performance in detecting non-muscle-invasive bladder cancer.

In any studies reporting the above outcomes, the following outcomes were also considered if reported:

• altered treatment as a result of the tests

• acceptability of the tests

• interpretability of the tests

• quality of life (disease-specific and generic instruments)

• adverse effects.

Exclusion criteria

The following types of report were excluded:

• animal models

• preclinical and biological studies

• reviews, editorials and opinions

• case reports

• abstracts, as usually insufficient methodological details are reported to allow critical appraisal of study quality

• reports investigating technical aspects of a test

• non-English language studies.

In addition, studies reporting biomarkers or cytology in which the number of participants in the analysis was less than 100 were excluded. Studies reporting cytology that predated the publication year of the earliest of the included biomarker studies were also excluded.

Diagnostic accuracy of PDD/urine biomarker tests

The results of the individual studies were tabulated and sensitivity, specificity, positive and negative predictive values, positive and negative likelihood ratios and diagnostic odds ratios (DORs) calculated. If reported in a given study, a separate 2 × 2 table was derived for patient-level and biopsy-level analyses.

Sensitivity describes the proportion of those with disease who have positive test results, whereas specificity is the proportion of those without disease who have negative test results. A positive likelihood ratio describes how many times more likely it is that a person with disease will receive a positive test result than a person without disease whereas a negative likelihood ratio describes how many times more likely it is that a person with disease will receive a negative test result than a person without disease. A positive predictive value (PPV) describes the proportion of those with positive test results who have the disease, whereas a negative predictive value (NPV) is the proportion of those with negative test results who do not have the disease. A DOR is a single indicator of test performance and is the ratio of the odds of testing positive in those with the disease relative to the odds of testing positive in those without the disease. It can be calculated from the sensitivity and specificity values. The DOR summarises the results into a single indicator of test performance; however, information contained in sensitivity and specificity is lost and in particular a DOR cannot distinguish between tests with high sensitivity and low specificity and vice versa.

Hierarchical summary receiver operating characteristic (HSROC) curves were produced for each test when three or more studies reported sufficient data. A separate HSROC curve was derived for patient-level analysis and biopsy-level analysis when possible. Meta-analysis models were fitted using the HSROC model48 in SAS 9.1 using the NLMIXED function (SAS Institute). This HSROC model takes account of the diseased and non-diseased sample sizes in each study and allows estimation of random effects for the threshold and accuracy effects. 48,49 HSROC models for PDD and WLC were fitted individually based upon the data for the individual alone, which allowed for an asymmetric summary receiver operating characteristic (SROC) curve. Additionally, two models that fitted the data simultaneously were also run, to formally assess the evidence for a difference in diagnostic accuracy between the tests. A fuller model was run that allowed for a difference between the tests in all three constituent diagnostic accuracy parameters (threshold, accuracy and shape of SROC curve) and also a simpler nested model was run that did not allow for a difference in diagnostic accuracy in any of the three parameters. The SROC curves from the HSROC models were produced and are shown on the corresponding SROC plots along with the individual study estimates. Summary sensitivity, specificity, positive and negative likelihood ratios and DORs for each model were reported as point estimate and 95% confidence interval (CI).

The presentation of test performance in terms of the detection of stage and grade of non-muscle-invasive bladder cancer was considered in the two broad categories of: (1) less aggressive, lower risk tumours (pTa, G1, G2) and (2) more aggressive, higher risk tumours (pT1, G3, CIS). The median (range) sensitivity of PDD and WLC across studies, for both patient- and biopsy-based detection of tumours, was reported for each category and also separately for CIS.

WLC-assisted TURBT compared with PDD-assisted TURBT

For relevant outcomes (e.g. recurrence rate after WLC-assisted TURBT compared with PDD-assisted TURBT), when appropriate, meta-analysis was employed to estimate a summary measure of effect. The dichotomous outcome data were combined using the Mantel–Haenszel (RR) method. For the estimates of RR, 95% CIs and p-values were calculated. The results were reported using a fixed-effect model in the absence of statistical heterogeneity. Chi-squared tests and I² statistics were used to explore statistical heterogeneity across studies. Possible reasons for heterogeneity were explored using sensitivity analysis. When there was no obvious reason for heterogeneity, the results were reported using random-effects methods. In the event that a quantitative synthesis was considered to be inappropriate or not feasible, we provided a narrative synthesis of results.

Overview

This section reports the diagnostic accuracy of PDD compared with WLC against a reference standard of histological assessment of biopsied tissue for the detection of non-muscle-invasive bladder cancer. The following levels of analysis are presented: patient, biopsy, stage/grade and photosensitising agent used. For patient and biopsy levels of analysis figures are included showing the sensitivity and specificity of the individual studies, SROC curves and pooled estimates with 95% CIs for sensitivity, specificity, positive and negative likelihood ratios and DORs for PDD and WLC. For the stage/grade level of analysis the median (range) sensitivity and specificity across studies are presented for PDD and WLC. Appendix 8 shows the studies that reported sufficient information (true and false positives and negatives for both PDD and WLC) to allow their inclusion in the pooled estimates for patient- and biopsy-level analysis, and also those studies comparing PDD with WLC that reported the sensitivity of the tests in detecting tumour stage/grade. Individual study results are given in Appendix 9. The results of studies reporting sensitivity and specificity for PDD but not WLC were examined to assess whether they differed from those of the comparative studies.

Patient-level analysis

Although biopsy-level analysis is useful to validate the accuracy of the test, patient-level data are more useful in determining management. Five studies65,66,73,80,81 comparing PDD with WLC and enrolling 386 people, with 370 included in the analysis, provided sufficient information to allow their inclusion in the pooled estimates for patient-level analysis. In four studies,65,66,80,81 of 318 patients included in the analysis, 131 (41%) had symptoms suggestive of bladder cancer and 187 (59%) had a history of non-muscle-invasive bladder cancer. The study by Riedl and colleagues73 did not report this information. Three of the studies65,66,81 used HAL as the photosensitising agent and two73,80 used 5-ALA. In two65,73 of the studies random biopsies of normal-appearing areas were taken.

Figure 9 shows the sensitivity and specificity of the individual studies, SROC curves and pooled estimates for the sensitivity and specificity of PDD and WLC for patient-based detection of bladder cancer. The pooled sensitivity (95% CI) for PDD was 92% (80% to 100%) compared with 71% (49% to 93%) for WLC, whereas the pooled specificity (95% CI) for PDD was 57% (36% to 79%) compared with 72% (47% to 96%) for WLC. The pooled estimates show that PDD had higher sensitivity but lower specificity than WLC, with the CIs for the two techniques overlapping. None of the five studies comparing PDD with WLC reported test performance separately for the group of patients newly presenting with a suspicion of bladder cancer or for the group with a history of non-muscle-invasive disease. The DOR values (95% CI) were 16.50 (1.00 to 42.23) for PDD and 6.44 (1.00 to 14.24) for WLC, with higher DORs indicating a better ability of the test to differentiate between those with and those without bladder cancer. Across studies the median (range) PPVs were 91% (59% to 100%) for PDD and 89% (56% to 100%) for WLC, and NPVs were 60% (32% to 100%) for PDD and 23% (20% to 87%) for WLC. However, it should be noted that predictive values are affected by disease prevalence, which is rarely constant across studies, and therefore these data should be interpreted with caution.

FIGURE 9.

Patient-level analysis: sensitivity, specificity, SROC curve and pooled estimates.

Three studies72,77,78 enrolling and analysing 153 patients reported patient-based detection for PDD only and were not included in the pooled estimates. All three studies used 5-ALA and, in one,72 random biopsies of normal-appearing areas were taken. Across these three studies the median (range) sensitivity and specificity for PDD were 91% (64% to 100%) and 67% (36% to 67%) respectively. In two72,78 of the studies the whole patient populations (n = 102) had a suspicion of bladder cancer with no previous history of the disease. Landry and colleagues72 reported a sensitivity of 64% for PDD, compared with 91% reported by Szygula and colleagues,78 whereas both studies reported a specificity of 67%.

Studies that reported patient-level analysis but only for CIS are considered in the section on stage/grade analysis.

Biopsy-level analysis

A total of 14 studies50,53,54,56,59–63,65,70,76,81,85 comparing PDD with WLC and enrolling 1751 people, with 1746 included in the analysis, provided sufficient information to allow their inclusion in the pooled estimates for biopsy-level analysis (number of biopsies: 8574 for PDD analysis, 8473 for WLC analysis). In nine studies,53,59–63,65,70,81 involving 1408 people, 560 (40%) had symptoms suggestive of bladder cancer and 848 (60%) had a history of non-muscle-invasive bladder cancer. The studies by Cheng and colleagues,50 Ehsan and colleagues,54 Filbeck and colleagues,56 Sim and colleagues76 and Zumbraegel and colleagues85 did not report this information. Ten studies50,53,54,56,59,61–63,70,85 used 5-ALA as the photosensitising agent and three60,65,81 used HAL, while the study by Sim and colleagues76 used hypericin. In eight studies50,53,54,59,60,63,65,70 random biopsies of normal-appearing areas were taken.

Figure 10 shows the sensitivity and specificity of the individual studies, SROC curves and pooled estimates for the sensitivity and specificity of PDD and WLC for biopsy-level detection of bladder cancer. In the pooled estimates, PDD had higher sensitivity (93%, 95% CI 90% to 96%) than WLC (65%, 95% CI 55% to 74%), whereas WLC had higher specificity (81%, 95% CI 73% to 90%) than PDD (60%, 95% CI 49% to 71%). The pair of CIs for both sensitivity and specificity did not overlap, providing evidence of a difference in diagnostic performance between the techniques. Across the 14 studies the sensitivity for PDD ranged from 76%65 to 98%54,62,70 compared with 1753 to 88%60 for WLC, and specificity ranged from 32%85 to 100%81 for PDD compared with 4685 to 100%81 for WLC. In the pooled analysis the DOR values (95% CI) were 20.29 (9.20 to 31.37) for PDD and 7.76 (3.39 to 11.93) for WLC. Across studies the median (range) PPVs were 61% (40% to 100%) for PDD and 70% (38% to 100%) for WLC, and the median (range) NPVs were 92% (20% to 99%) for PDD and 78% (13% to 91%) for WLC.

FIGURE 10.

Biopsy-level analysis: sensitivity, specificity, SROC curve and pooled estimates.

None of the 14 studies comparing PDD with WLC reported biopsy-level detection separately for the group of patients newly presenting with a suspicion of bladder cancer or for the group with a history of non-muscle-invasive disease.

Six studies58,67,71,73,77,84 involving 428 patients reported biopsy-level detection for PDD only and were not included in the pooled estimates. All six studies used 5-ALA and in four58,67,73,84 random biopsies of normal-appearing areas were taken. Across the six studies the median (range) sensitivity and specificity for PDD were 95% (87% to 98%) and 51% (36% to 67%) respectively. In two58,84 of these studies the whole patient population (n = 68) had a history of non-muscle-invasive bladder cancer. Frimberger and colleagues58 and Zaak and colleagues84 reported sensitivities of 95% and 90% and specificities of 67% and 61%, respectively, for PDD.

Studies that reported biopsy-level analysis but only for CIS are included in the section on stage/grade analysis.

Formal comparison of PDD and WLC in patient- and biopsy-based analysis

In addition to the two HSROC models of the diagnostic accuracy of PDD and WLC individually, two HSROC models were run that simultaneously modelled PDD and WLC diagnostic accuracy using all of the data from the 14 studies. There was strong evidence of a difference in diagnostic accuracy between the tests, with the model that allowed for a difference in diagnostic accuracy in the three constituent parameters (threshold, accuracy and shape of SROC curve) having a substantially better Bayesian information criterion than the simplified diagnostic accuracy model, for both patient- and biopsy-level analysis (difference of 1408.0 and 20.7 respectively). These results are supported by noting that the intervals for the summary sensitivity and specificity at biopsy level from the models in which the tests were modelled separately (Figure 10) did not overlap for either measure. PDD had a greater sensitivity than WLC but at the cost of a lower specificity. The point estimates of the patient-level analysis were similar to those from the biopsy-level analysis, although the intervals were substantially wider, as might be expected because of the smaller number of studies and observations available for this level of analysis.

Stage/grade analysis

Studies reporting the sensitivity of PDD compared with WLC in the detection of stage and grade of tumour categorised this information in different ways, including pTa, pTaG1, pTaG1–2, pTaG2, pTaG2–3, pTaG3, pTa-T1, G1–2, pT1, pT1G1, pT1G1–2, pT1G2, pT1G3, > pT1, CIS, G3, pT2G2, pT2G3, ≥ pT2, ≥ pT2G3 and pT4G3 (see Appendix 8). Some studies reported the detection of stage/grade at the patient level and others reported this information at biopsy level.

For the purposes of this review, the presentation of test performance in terms of the detection of stage and grade of non-muscle-invasive bladder cancer was considered in two broad categories:

1. less aggressive, lower risk tumours (pTa, G1, G2)

2. more aggressive, higher risk tumours (pT1, G3, CIS).

Table 7 shows the median (range) sensitivity of PDD and WLC across studies, for both patient- and biopsy-based detection of tumours, within the broad categories of less aggressive/lower risk and more aggressive/higher risk (including CIS), and also separately for CIS.

Photosensitising agent used

Table 9 shows the median (range) sensitivity and specificity across studies for the different photosensitising agents used, for both patient- and biopsy-level detection of bladder cancer. Four studies using 5-ALA72,73,77,78 and three using HAL65,66,81 reported patient-level detection of bladder cancer. Across the studies using 5-ALA the median (range) sensitivity and specificity were 96% (64% to 100%) and 52% (33% to 67%), respectively, compared with 90% (53% to 96%) sensitivity and 81% (43% to 100%) specificity for HAL.

A total of 15 studies using 5-ALA,50,53,54,56,58,59,61,63,67,70,71,73,77,84,85 three using HAL60,65,81 and one using hypericin76 reported biopsy-level detection of bladder cancer. Across the studies using 5-ALA the median (range) sensitivity and specificity were 95% (87% to 98%) and 57% (32% to 67%), respectively, compared with 85% (76% to 94%) sensitivity and 80% (58% to 100%) specificity for HAL. The study by Sim and colleagues76 reported 82% sensitivity and 91% specificity for hypericin.

The results for both patient- and biopsy-based detection suggest that 5-ALA may have slightly higher sensitivity than HAL, whereas HAL may have higher specificity than 5-ALA, but this should be interpreted with caution as factors other than the photosensitising agent used may have contributed to the sensitivity and specificity values reported by the studies.

Four studies reported sensitivity and specificity at both patient and biopsy level, two using 5-ALA73,77 and two using HAL. 65,81

Side effects of photosensitising agents

Overview

This section presents the results of the four RCTs86,88,89,92 comparing PDD with WLC and reporting the effectiveness outcomes of recurrence-free survival, residual tumour rate at first cystoscopy following TURBT, recurrence rate during follow-up and tumour progression. Random-effects meta-analyses using RR as the effect measure are presented comparing PDD and WLC in terms of these outcomes.

The RCTs enrolled 709 participants, with 544 included in the analysis. In the study by Daniltchenko and colleagues88 the groups were randomised to WLC or PDD, whereas in the studies by Babjuk and colleagues,86 Denzinger and colleagues89 and Kriegmair and colleagues92 the groups were randomised to WLC or WLC and PDD. The follow-up periods varied from 10–14 days for the study by Kriegmair and colleagues,92 which evaluated residual tumour following TURBT, to 2 years for the study by Babjuk and colleagues,86 5 years for the study by Daniltchenko and colleagues88 and 8 years for the study by Denzinger and colleagues. 89 All four studies used 5-ALA as the photosensitising agent. Individual study results are given in Appendix 9.

In the study by Babjuk and colleagues86 none of the randomised patients with grade 1 or grade 2 tumours received adjuvant intravesical therapy during the study. All patients with grade 3 tumours (six in the PDD group and seven in the WLC group) received intravesical BCG immunotherapy, based on a standard 6-week course followed by three, weekly instillations (3-week course) at 3, 6 and 12 months. 86 In the study by Daniltchenko and colleagues88 none of the randomised patients received adjuvant intravesical therapy throughout the study. In the study by Denzinger and colleagues89 patients with a solitary primary tumour staged pTaG1–G2 did not receive recurrence prophylaxis. Patients with multifocal involvement of the bladder staged pTaG1–G2 or pT1G1–G2 underwent mitomycin therapy, and those with primary stage pT1G3, CIS or treatment failure with mitomycin received BCG therapy, with weekly instillations of 120 mg BCG given for 6 weeks. 89 The study by Kriegmair and colleagues92 did not state whether adjuvant intravesical therapy was given, although the primary outcome of this study was to evaluate residual tumour 10–14 days following TURBT.

The four RCTs were reported in eight reports. The study for which Denzinger and colleagues89 is considered the primary report was also reported by Filbeck and colleagues,91 Burger and colleagues87 and Denzinger and colleagues. 90 The primary report gave information on recurrence-free survival at 2, 4, 6 and 8 years and also tumour recurrence throughout this follow-up period, overall and for low-, intermediate- and high-risk groups, as well as reporting residual tumour rate at secondary transurethral resection (TUR). Filbeck and colleagues91 reported residual tumour rate 6 weeks after initial resection and recurrence-free survival at 12 and 24 months. Burger and colleagues87 reported recurrence-free survival, and tumour recurrence and progression at 7.1 years, and Denzinger and colleagues90 reported recurrence-free survival and tumour recurrence and progression for a subgroup of patients who presented with initial T1 high-grade bladder cancer.

The study for which Daniltchenko and colleagues88 is considered the primary report was also reported by Riedl and colleagues. 93 Daniltchenko and colleagues88 reported tumour recurrence and progression during follow-up whereas Riedl and colleagues reported residual tumour rate at the control TUR. 93

Recurrence-free survival

The studies by Babjuk and colleagues86 and Denzinger and colleagues89 involving a total of 313 patients reported recurrence-free survival at 12 and 24 months. In a random-effects meta-analysis comparing PDD and WLC in terms of recurrence-free survival, the direction of effect of the pooled estimate at both time points favoured PDD over WLC, although the difference was statistically significant only at 24 months (Figure 11). There was evidence of substantial statistical heterogeneity between the studies at the 12-month time point (I² = 74.4%).

FIGURE 11.

Recurrence-free survival.

Denzinger and colleagues90 also reported on a subgroup of 46 patients who were diagnosed with T1 high-grade bladder cancer, with recurrence-free survival rates of 80% (17/21) in the PDD group compared with 52% (13/25) in the WLC group at the 8-year follow-up.

Residual tumour rate at first cystoscopy following transurethral resection

The studies by Babjuk and colleagues,86 Daniltchenko and colleagues,88 Denzinger and colleagues89 and Kriegmair and colleagues92 involving a total of 534 patients reported residual tumour rate at first cystoscopy following TUR. The timing of the cystoscopy varied between the studies, with Kriegmair and colleagues92 reporting the residual tumour rate 10–14 days after the initial resection, Denzinger and colleagues89 and Riedl and colleagues93 reporting it 6 weeks after initial resection, and Babjuk and colleagues86 assessing the residual tumour rate 10–15 weeks after TUR.

Figure 12 shows a random-effects meta-analysis comparing PDD with WLC in terms of residual tumour (pTa and pT1) detected at first cystoscopy following the initial TUR. The pooled estimates show that PDD resulted in both statistically significantly fewer residual pTa tumours (RR 0.32, 95% CI 0.15 to 0.70) and fewer pT1 tumours (RR 0.26, 95% CI 0.12 to 0.57), with an overall RR of 0.37 (95% CI 0.20 to 0.69) in favour of PDD (Kriegmair and colleagues92 reported overall rates only).

FIGURE 12.

Residual tumour (pTa and pT1) at first cystoscopy following TUR. Notes: 1. In the figure, the numbers of patients shown for the study by Denzinger and colleagues89 do not include five each from the PDD and WLC groups who at initial resection had CIS. At 6 weeks after initial resection none of the five patients in the PDD group were found to have residual CIS but four of five (80%) in the WLC group were found to have residual CIS. 2. Kriegmair and colleagues92 reported that in an intention to treat analysis 61.5% (40/65) of patients in the PDD group and 40.6% (26/64) of patients in the WLC group were tumour free. For the purposes of the meta-analysis this was interpreted as 25/64 patients in the PDD group and 38/64 patients in the WLC group having residual tumour.

The studies by Babjuk and colleagues86 and Daniltchenko and colleagues88 also reported residual tumour according to grade, and Figure 13 shows a fixed-effect meta-analysis comparing PDD with WLC in terms of the grade of residual tumour detected at first cystoscopy following the initial TUR. The pooled estimates for G3 were not statistically significant, whereas those for G1 (RR 0.13, 95% CI 0.03 to 0.71), G2 (RR 0.32, 95% CI 0.16 to 0.64) and overall (RR 0.31, 95% CI 0.18 to 0.53) showed a statistically significant difference in favour of PDD. In the study by Babjuk and colleagues86 none of the patients with grade 1 or grade 2 tumours received adjuvant intravesical therapy whereas all those with grade 3 tumours received intravesical BCG immunotherapy. In the study by Daniltchenko and colleagues88 none of the patients received adjuvant intravesical therapy.

FIGURE 13.

Residual tumour (G1, G2 and G3) at first cystoscopy following transurethral resection.

Tumour recurrence rate during follow-up

The studies by Daniltchenko and colleagues88 and Denzinger and colleagues89 involving a total of 293 patients reported tumour recurrence rate during the follow-up period. The follow-up period for the study by Daniltchenko and colleagues was 5 years88 whereas that for the study by Denzinger and colleagues was 8 years. 89

Figure 14 shows a random-effects meta-analysis comparing PDD with WLC in terms of the number of patients who experienced tumour recurrence during the follow-up period. Although the direction of effect for both studies favoured PDD it was statistically significant only in the study by Denzinger and colleagues, and the pooled estimate did not show a statistically significant difference between PDD and WLC (RR 0.64, 95% CI 0.39 to 1.06). 89 There was evidence of substantial statistical heterogeneity between the studies (I² = 71.1%).

FIGURE 14.

Tumour recurrence rates during the follow-up period.

In the study by Daniltchenko and colleagues88 none of the randomised patients received adjuvant intravesical therapy. In the study by Denzinger and colleagues89 patients with a solitary primary tumour staged pTaG1–G2 (low-risk group) did not receive adjuvant intravesical therapy. Patients with multifocal involvement of the bladder staged pTaG1–G2 or pT1G1–G2 (intermediate-risk group) underwent mitomycin therapy, and those with primary stage pT1G3, CIS or treatment failure with mitomycin (high-risk group) received BCG therapy, with weekly instillations of 120 mg BCG given for 6 weeks. 89 Table 10 shows the recurrence rates for the low-, intermediate- and high-risk groups over the 8-year follow-up in the study by Denzinger and colleagues. 89 Although there were consistently fewer recurrences for PDD compared with WLC across all risk groups, the difference in recurrence rates between PDD and WLC was smaller in the intermediate- and high-risk groups, both of which received adjuvant intravesical therapy, albeit with wide CIs.

In the subgroup of 46 patients initially diagnosed with T1 high-grade bladder cancer, Denzinger and colleagues89 reported recurrence rates of 14% (3/21) in the PDD group compared with 44% (11/25) in the WLC group during the follow-up period.

Time to recurrence

The studies by Babjuk and colleagues86 and Daniltchenko and colleagues88 reported time to recurrence of bladder tumours. In the study by Babjuk and colleagues86 this was a median of 17.05 months for the PDD group and 8.05 months for the WLC group. Babjuk and colleagues86 also reported a median time to recurrence in patients with multiple tumours of 13.54 months for the PDD group and 4.45 months for the WLC group. Daniltchenko and colleagues88 reported a median (range) time to recurrence of 12 months (2 to 58) for the PDD group and 5 months (2 to 52) for the WLC group.

Tumour progression during follow-up

The studies by Daniltchenko and colleagues88 and Denzinger and colleagues89 also reported tumour progression during their follow-up periods of 5 years and 8 years respectively.

Figure 15 shows a fixed-effect meta-analysis comparing PDD with WLC in terms of the numbers of patients who experienced tumour progression during the follow-up period. The direction of effect of the study by Daniltchenko and colleagues88 favoured PDD (four versus nine events) whereas in the study by Denzinger and colleagues89 there were two cases in each group. The pooled estimate had wide CIs reflecting the small number of events (RR 0.57, 95% CI 0.22 to 1.46).

FIGURE 15.

Tumour progression rates during the follow-up period.

In the subgroup of patients diagnosed with T1 high-grade bladder cancer, Denzinger and colleagues90 reported progression to muscle-invasive disease (≥ T2) of 19% (4/21) in the PDD group compared with 12% (3/25) in the WLC group during the follow-up period.

Assessment of diagnostic accuracy

A total of 31 studies, published in 44 reports, met the inclusion criteria for the PDD part of the review. In total, 27 studies (36 reports) reported the diagnostic accuracy of PDD. As measured by the modified QUADAS checklist, in all studies partial verification bias was avoided (all patients received a reference standard test) and test review bias was avoided (PDD and WLC were interpreted without knowledge of the results of the reference standard test). In 96% (26/27) of studies uninterpretable or intermediate test results were reported or there were none, and withdrawals from the study were explained or there were none. However, all of the studies were judged to suffer from incorporation bias in that PDD was considered not to be independent of the reference standard test as biopsies used in the reference standard test were obtained via the PDD procedure.

In both patient- and biopsy-based detection of bladder cancer PDD had higher sensitivity but lower specificity than those of WLC. Five studies involving 370 patients reported patient-based detection. In the pooled estimates the sensitivity for PDD was 92% (95% CI 80% to 100%) compared with 71% (95% CI 49% to 93%) for WLC, whereas the specificity for PDD was 57% (95% CI 36% to 79%) compared with 72% (95% CI 47% to 96%) for WLC, with the CIs for the two techniques overlapping. A total of 14 studies involving 1746 patients reported biopsy-based detection (number of biopsies: 8574 for PDD analysis, 8473 for WLC analysis). In the pooled estimates the sensitivity for PDD was 93% (95% CI 90% to 96%) compared with 65% (95% CI 55% to 74%) for WLC, whereas the specificity for PDD was 60% (95% CI 49% to 71%) compared with 81% (95% CI 73% to 90%) for WLC. The pair of CIs for both sensitivity and specificity did not overlap, providing evidence of a difference in diagnostic performance between the techniques.

Studies reporting the sensitivity of PDD compared with WLC for detecting stage/grade of bladder cancer categorised this information in different ways. For the purposes of this review the detection of stage/grade was considered in two broad categories:

1. less aggressive, lower risk tumours (pTa, G1, G2)

2. more aggressive, higher risk tumours (pT1, G3, CIS).

Across three studies66,80,81 involving 266 patients reporting patient-based detection of lower risk, less aggressive tumours, the median (range) sensitivity of PDD at 92% (20% to 95%) was broadly similar to that of WLC at 95% (8% to 100%). Across seven studies54,56,60–62,67,81 involving 1206 patients reporting biopsy-based detection (n = 5777 biopsies overall), the median (range) sensitivity of PDD was slightly higher at 96% (88% to 100%) compared with 88% (74% to 100%) for WLC. Across six studies51,57,65,66,80,81 involving 563 patients reporting patient-based detection of more aggressive, higher risk tumours, the median (range) sensitivity of PDD at 89% (6% to 100%) was higher than that of WLC at 56% (0% to 100%). Across 13 studies50,53,54,56,57,60–62,65,67,70,81,85 involving 1756 patients reporting biopsy-based detection (n = 7506 biopsies overall), the median (range) sensitivity of PDD at 99% (54% to 100%) was again much higher than that of WLC at 67% (0% to 100%) (Table 7). These results suggest that PDD is much better than WLC in detecting more aggressive, higher risk tumours. However, the results for patient- and biopsy-based detection for less aggressive, lower risk tumours and patient-based detection for more aggressive, higher risk tumours should be interpreted with caution as they are based on only a small number of studies.

When CIS was considered separately, across six studies51,57,65,66,80,81 involving 563 patients reporting patient-based detection, the median (range) sensitivity of PDD for detecting CIS at 83% (41% to 100%) was much higher than that of WLC at 32% (0% to 83%). Across 13 studies50,53,54,56,57,60–62,65,67,70,81,85 involving 1756 patients reporting biopsy-based detection of CIS (n = 7506 biopsies overall), the median (range) sensitivity of PDD at 86% (54% to 100%) was also much higher than that of WLC at 50% (0% to 68%). The results for patient-based detection should be interpreted with caution as they are based on only a small number of studies. However, the median sensitivity across studies reported for patient-based detection of CIS (83%) was similar to that reported for biopsy-based detection of CIS (86%). Only three studies reported the specificity of PDD and WLC for detecting CIS. Two studies51,70 reported higher specificity for WLC (97% versus 71% and 68% versus 61% respectively), whereas the third57 reported similar specificity for both techniques (83% for WLC versus 82% for PDD).

Of the studies comparing PDD with WLC that were included in the pooled estimates in the present review, two65,73 of five reporting patient-based analysis and eight50,53,54,59,60,63,65,70 of 14 reporting biopsy-based analysis undertook random biopsies of normal-appearing areas. Ten50,53,54,56,60–62,70,81,85 of these 14 studies also reported detection of CIS lesions. Table 11 shows, for patient- and biopsy-level analysis and also for detection of CIS lesions, the sensitivity and specificity for PDD and WLC for those studies included in the pooled estimates that undertook random biopsies compared with those that did not. There did not appear to be any systematic pattern to the performance of the tests based on whether or not random biopsies were undertaken.

Most studies (n = 18) used 5-ALA as the photosensitising agent, with five using HAL, two hypericin and two either 5-ALA or HAL. In patient-based detection of bladder cancer, across four studies using 5-ALA72,73,77,78 and three using HAL,65,66,81 the median (range) sensitivity and specificity for 5-ALA were 96% (64% to 100%) and 52% (33% to 67%), respectively, compared with 90% (53% to 96%) sensitivity and 81% (43% to 100%) specificity for HAL. In biopsy-based detection of bladder cancer, across 15 studies50,53,54,56,58,59,61,63,67,70,71,73,77,84,85 using 5-ALA, the median (range) sensitivity and specificity for 5-ALA were 95% (87% to 98%) and 57% (32% to 67%), respectively, compared with 85% (76% to 94%) and 80% (58% to 100%) for HAL. One study, by Sim and colleagues,76 used hypericin, reporting 82% sensitivity and 91% specificity. The results for both patient- and biopsy-based detection suggest that 5-ALA may have slightly higher sensitivity than HAL, whereas HAL may have higher specificity than 5-ALA, but this should be interpreted with caution as factors other than the photosensitising agent used may have contributed to the sensitivity and specificity values reported by the studies.

In total, 20 studies reported side effects. Twelve studies51–53,61–63,65,71–73,78,81 involving 1543 patients reported that there were no side effects or no serious side effects associated with the photosensitising agent used (5-ALA, eight studies; HAL, two studies; 5-ALA/HAL not reported separately, one study; hypericin, one study). In four studies50,67,71,77 involving 245 patients and using 5-ALA, reported side effects associated with the agent included nine patients who complained of urgency,50,70 four with alginuresis symptoms and pollakiuria,70 three with significant gram-negative bacteriuria,70 one with acute cystitis accompanied by haemorrhagic lesion,77 one with transient dysuria67 and one who developed a urinary tract infection. 67 Two studies57,66 involving 460 patients and using HAL reported 21 non-serious side effects that were associated with the agent. One study76 involving 41 patients and using hypericin reported that one patient developed microscopic haematuria from cystitis.

In summary, the evidence suggests that PDD has clinically important better sensitivity but lower specificity than WLC in the detection of bladder cancer and, in terms of stage/grade, has higher sensitivity than WLC in the detection of more aggressive, higher risk tumours (pT1, G3, CIS).

Assessment of recurrence/progression of disease

Four RCTs (eight reports) reporting recurrence/progression enrolled 709 participants, with 544 included in the analysis. The follow-up periods varied from 10–14 days for the study by Kriegmair and colleagues92 (although the aim of this study was to evaluate residual tumour following TURBT) to 2 years for the study by Babjuk and colleagues,86 5 years for the study by Daniltchenko and colleagues88 and 8 years for the study by Denzinger and colleagues. 89 All four studies used 5-ALA as the photosensitising agent.

The study by Daniltchenko and colleagues88 reported that none of the patients received adjuvant intravesical therapy. In the study by Babjuk and colleagues86 only patients with grade 3 tumours received intravesical therapy. In the study by Denzinger and colleagues89 patients with a solitary primary tumour staged pTaG1–G2 did not receive intravesical therapy, whereas those with multifocal tumours staged pTaG1–G2 or pT1G1–G2 underwent mitomycin therapy and those with primary stage pT1G3, CIS or treatment failure with mitomycin received BCG therapy. The study by Kriegmair and colleagues92 did not state whether intravesical therapy was given.

In all four studies the PDD and WLC groups were similar at baseline in terms of prognostic factors, eligibility criteria for the studies were specified, and length of follow-up was considered adequate in relation to the outcomes of interest reported by the studies. However, in all four studies it was unclear whether the sequence generation was really random or whether the treatment allocation was adequately concealed.

The studies by Babjuk and colleagues86 and Denzinger and colleagues89 (involving a total of 313 patients) reported recurrence-free survival at 12 and 24 months. In a random-effects meta-analysis the direction of effect of the pooled estimate at both time points favoured PDD over WLC, although the difference was statistically significant only at 24 months (RR 1.37, 95% CI 1.18 to 1.59).

The studies by Babjuk and colleagues,86 Daniltchenko and colleagues,88 Denzinger and colleagues89 and Kriegmair and colleagues92 involving a total of 534 patients reported residual tumour rate at first cystoscopy following TURBT. In a random-effects meta-analysis PDD was associated with both statistically significantly fewer residual pTa tumours (RR 0.32, 95% CI 0.15 to 0.70) and pT1 tumours (RR 0.26, 95% CI 0.12 to 0.57), with an overall pooled estimate RR of 0.37 (95% CI 0.20 to 0.69) in favour of PDD. Babjuk and colleagues86 and Daniltchenko and colleagues88 also reported residual tumour according to grade (G1, G2 and G3). In a fixed-effect meta-analysis the pooled estimates for G1 (RR 0.13, 95% CI 0.03 to 0.71) and G2 (RR 0.32, 95% CI 0.16 to 0.64) were statistically significant in favour of PDD, as was the overall pooled estimate (RR 0.31, 95% CI 0.18 to 0.53).

Daniltchenko and colleagues88 and Denzinger and colleagues,89 in studies involving a total of 293 patients, reported tumour recurrence rate during the follow-up period (5 years and 8 years respectively). In a random-effects meta-analysis of the number of patients who experienced tumour recurrence, although the direction of effect for both studies favoured PDD it was statistically significant only in the study by Denzinger and colleagues,89 and the direction of effect in the pooled estimate also favoured PDD but was not statistically significant (RR 0.64, 95% CI 0.39 to 1.06). In the study by Denzinger and colleagues89 the recurrence rates were consistently lower for PDD than for WLC across all three risk groups. However, the difference in the recurrence rates between PDD and WLC was smaller in the intermediate-risk [PDD 7% (6/88), WLC 13% (13/103)] and high-risk [PDD 7% (6/88), WLC 10% (10/103)] groups that received adjuvant intravesical therapy than in the low-risk group that did not [PDD 7% (6/88), WLC 19% (20/103)].

Two studies86,88 reported time to recurrence, both favouring PDD. Babjuk and colleagues86 reported a median time to recurrence of 17.05 months for the PDD group and 8.05 months for the WLC group, whereas Daniltchenko and colleagues88 reported a median (range) time to recurrence of 12 (2 to 58) months for the PDD group and 5 (2 to 52) months for the WLC group.

The studies by Daniltchenko and colleagues88 and Denzinger and colleagues89 also reported tumour progression during their respective 5- and 8-year follow-up periods. In a fixed-effect meta-analysis of the number of patients who experienced tumour progression, the direction of effect of the study by Daniltchenko and colleagues favoured PDD whereas that of the study by Denzinger and colleagues favoured WLC, although neither was statistically significant. The pooled estimate favoured PDD but again was not statistically significant (RR 0.57, 95% CI 0.22 to 1.46). 88,89

In summary, the evidence suggests that, compared with WLC, the use of PDD at TURBT results in less residual tumour being found at the first cystoscopy following TURBT, longer recurrence-free survival of patients and a longer time to recurrence following TURBT, and may be associated with a lower rate of tumour recurrence over time. However, as these results are based on only a few studies they should be interpreted with caution. It should also be borne in mind that the administration of adjuvant intravesical therapy varied across the studies. Adjuvant intravesical therapy following TURBT is standard practice in the UK and much of Europe and can reduce recurrence by up to 50% in the first 2 years. The fact that in two studies86,89 only some patients received intravesical therapy and in one88 none did, while in the fourth study92 this information was not reported, makes it difficult to assess what the true added value of PDD might be in reducing recurrence rates in routine practice.

Characteristics of the included studies

A description of each of the 14 included FISH studies is given in Appendix 12, which contains the characteristics of all of the included biomarker and cytology studies listed alphabetically by surname of first author. Table 12 shows summary information for the 14 FISH studies.

Twelve studies, reported in 13 papers,94–106 were diagnostic cross-sectional studies, of which two101,104 reported consecutive recruitment, and the remaining two107,108 were case–control studies. Two studies were multicentre (21 centres,108 23 centres103).

The 14 studies enrolled 3321 participants, with 2961 included in the analysis. In 12 studies94,95,97,99,101–108 reporting this information, 1012 (45%) presented with a suspicion of bladder cancer and 1234 (55%) had previously diagnosed bladder cancer. In one103 of these studies the whole study population (n = 497) had a suspicion of bladder cancer and in two102,106 the whole study population had previously diagnosed bladder cancer (n = 355). Two studies98,100 did not report this information.

Across seven studies97,99,101–103,106,107 providing information on patient age for the whole study population, the median (range) of means/medians was 70 years (63 to 72 years) (Yoder and colleagues106 reported median rather than mean age). Seven studies94,97,99,102,103,106,107 provided information on the gender of 1512 participants, of whom 1073 (71%) were men and 439 (29%) were women.

Seven studies94,99,102–104,106,108 gave details of when they took place, with an earliest start date of 1996104 and latest end date of March 2007. 99 Seven studies took place in the USA,97,99,103–106,108 three in Germany95,98,107 and one each in the Netherlands102 and Israel,94 and two had multinational settings, taking place in Austria/Italy101 and the USA/Belgium. 100

Methodological quality of the included studies

Figure 17 summarises the quality assessment across the 14 FISH studies. The results for the quality assessment of the individual studies are shown in Appendix 13. In all studies the spectrum of patients who received the tests was considered to be representative of those who would receive the test in practice (Q1). For this question we considered patients to be representative if the patient population either had a suspicion or a history of bladder cancer or contained patients from both groups, or the majority or all of the patient population presented with either gross or microhaematuria or contained a mixture of patients with either indication. In all studies the reference standard (cystoscopy with histological assessment of biopsied tissue) was considered likely to correctly classify bladder cancer (Q2). In all studies partial verification bias was avoided in that all patients who underwent a FISH test also received a reference standard test (Q4), differential verification bias was avoided in that patients received the same reference standard regardless of the index test result (Q5) and incorporation bias was avoided in that the reference standard was independent of the index test (Q6). In all studies either uninterpretable or intermediate test results were reported or there were none (Q10), and withdrawals from the study were explained or there were none (Q11).

FIGURE 17.

Summary of quality assessment of FISH studies (n = 14).

In 10 studies (71%) the time period between FISH and the reference standard was considered to be short enough (1 month or less) to be reasonably sure that the patient’s condition had not changed in the intervening period (Q3). In nine studies (64%) test review bias was avoided in that the FISH results were interpreted without knowledge of the results of the reference standard test (Q7). However, in nine studies (64%) it was unclear whether the reference standard results were interpreted without knowledge of the results of the FISH test (diagnostic review bias, Q8) and in eight studies (57%) it was unclear whether the same clinical data were available when test results were interpreted as would be available when the test is used in practice (clinical review bias, Q9). In this context clinical data were defined broadly to include any information relating to the patient such as age, gender, presence and severity of symptoms, and other test results.

In 13 studies (93%) a prespecified cut-off value was used (Q12); in 10 studies (71%) a clear definition of what was considered to be a positive test result was provided (Q13); and none of the studies provided information on observer variation in interpretation of test results (Q14).

Assessment of diagnostic accuracy

Patient-level analysis

A total of 12 studies95,97–101,103–108 enrolling 3101 people, with 2535 included in the analysis, provided sufficient information to allow their inclusion in the pooled estimates for patient-level analysis. The cut-offs used by these studies to define a positive test result were considered sufficiently similar for all of them to be included in the pooled estimates (see Appendix 16 for a description of the cut-offs used by each of the FISH studies). Figure 18 shows the sensitivity and specificity of the individual FISH studies, pooled estimates and SROC curve for patient-based detection of bladder cancer. The pooled sensitivity and specificity (95% CI) were 76% (65% to 84%) and 85% (78% to 92%), respectively, and the DOR (95% CI) value was 18 (3 to 32). Across the 12 studies the sensitivity for FISH ranged from 53%107 to 96%,101 and specificity ranged from 45%101 to 97%. 104 The median (range) PPV across studies was 78% (27% to 99%) and the median (range) NPV was 88% (36% to 97%). However, as previously mentioned, predictive values are affected by disease prevalence, which is rarely constant across studies, and therefore these data should be interpreted with caution.

FIGURE 18.

FISH patient-level analysis: sensitivity, specificity, pooled estimates and SROC curve.

Most of the included studies in the pooled estimates contained a mixture of patients with a suspicion of bladder cancer and those with previously diagnosed bladder cancer, although two studies98,100 did not report this information. In the study by Sarosdy and colleagues103 all of the participants (n = 497) had a suspicion of bladder cancer (sensitivity 69%, specificity 78%) and in the study by Yoder and colleagues106 all of the participants (n = 250) had previously diagnosed bladder cancer (sensitivity 64%, specificity 73%).

Characteristics of the included studies

A description of each of the 10 included ImmunoCyt studies is given in Appendix 12, which contains the characteristics of all of the included biomarker and cytology studies listed alphabetically by surname of first author. Table 14 shows summary information for the 10 ImmunoCyt studies.

All 10 studies, reported in 12 papers,101,109–119 were diagnostic cross-sectional studies. Six reported consecutive recruitment. 101,109,110,112,116,118 Two studies were multicentre (four centres,111 19 centres116).

The 10 studies enrolled 4199 participants, with at least 3091 included in the analysis (the study by Mian and colleagues113 enrolled 942 participants but did not report the number included in the analysis). In nine studies101,109–114,116,118 reporting this information, 890 participants (27%) presented with a suspicion of bladder cancer and 2405 (73%) had previously diagnosed bladder cancer. In one of these studies118 the whole patient population (n = 301) had a suspicion of bladder cancer and in three110,111,113 the whole population had previously diagnosed bladder cancer (n = 1499). One study119 did not report this information.

Across six studies101,109,112–114,116 providing information on patient age for the participant group as a whole, the median (range) of means was 68 years (66 to 73 years). Four studies112,114,116,118 provided information on the gender of 1371 participants, of whom 1076 (78%) were men and 295 (22%) were women.

Six studies111–114,118,119 gave details of when they took place, with an earliest start date of November 1997112 and latest end date of July 2007. 118 The studies took place in Austria,112 France,116 Germany,118 Italy,113 Sweden,114 Canada119 and the USA,111 with three having multinational settings, all taking place in Austria/Italy,101,109,110 although they did not state that they were multicentre.

Methodological quality of the included studies

Figure 19 summarises the quality assessment for the 10 ImmunoCyt studies. The results for the quality assessment of the individual studies are shown in Appendix 13. In all studies the spectrum of patients who received the tests was considered to be representative of those who would receive the test in practice (Q1). In all studies the reference standard (cystoscopy with histological assessment of biopsied tissue) was considered likely to correctly classify bladder cancer (Q2). In all studies partial verification bias was avoided in that all patients who underwent an ImmunoCyt test also received a reference standard test (Q4), differential verification bias was avoided in that patients received the same reference standard regardless of the index test result (Q5) and incorporation bias was avoided in that the reference standard was independent of the index test (Q6). In all studies either uninterpretable or intermediate test results were reported or there were none (Q10) and a prespecified cut-off value was used (Q12). In eight studies (80%) the time period between ImmunoCyt and the reference standard was considered to be short enough (1 month or less) to be reasonably sure that the patient’s condition had not changed in the intervening period (Q3). In nine studies (90%) withdrawals from the study were explained or there were none (Q11) and a clear definition of what was considered to be a positive result was provided (Q13).

FIGURE 19.

Summary of quality assessment of ImmunoCyt studies (n = 10).

In all 10 studies (100%) it was unclear whether diagnostic review bias had been avoided (Q8), in nine studies (90%) it was unclear whether clinical review bias had been avoided (Q9) and in seven studies (70%) it was unclear whether test review bias had been avoided (Q7). One study (10%) provided information on observer variation in interpretation of test results (Q14).

Assessment of diagnostic accuracy

Patient-level analysis

Eight studies101,109,111,112,114,116,118,119 enrolling 3041 participants, with 2896 included in the analysis, provided sufficient information to allow their inclusion in the pooled estimates for patient-level analysis. The ‘common’ cut-off used by all of these studies to define a positive test result was at least one green or one red fluorescent cell. Figure 20 shows the sensitivity and specificity of the individual studies, pooled estimates and SROC curve for ImmunoCyt patient-based detection of bladder cancer. The pooled sensitivity and specificity (95% CI) were 84% (77% to 91%) and 75% (68% to 83%), respectively, and the DOR value (95% CI) was 16 (6 to 26). Across the studies the sensitivity for ImmunoCyt ranged from 73%116 to 100%,114 and specificity ranged from 62%119 to 88%. 118 The median (range) PPV across studies was 54% (26% to 70%) and the median (range) NPV was 93% (86% to 100%).

FIGURE 20.

ImmunoCyt patient-level analysis: sensitivity, specificity, pooled estimates and SROC curve.

Most of the included studies in the pooled estimates contained a mixture of patients with a suspicion of bladder cancer and those with previously diagnosed bladder cancer, although one study119 did not report this information. In the study by Schmitz-Drager and colleagues118 all of the participants (n = 280) had a suspicion of bladder cancer (sensitivity 85%, specificity 88%) and in the study by Messing and colleagues111 all of the participants (n = 326) had previously diagnosed bladder cancer (sensitivity 81%, specificity 75%).

Characteristics of the included studies

A description of each of the 41 included NMP22 studies is given in Appendix 12, which contains the characteristics of all of the included biomarker and cytology studies listed alphabetically by surname of first author. Table 16 shows summary information for the 41 NMP22 studies.

Thirty-one studies, reported in 37 papers,45,80,95,120–153 were diagnostic cross-sectional studies. Three45,126,127 reported consecutive recruitment. A total of 10 studies, reported in 11 papers,154–164 were case–control studies. Four studies were multicentre (23 centres,126 23 centres,127 13 centres,135 three centres149).

The 41 studies enrolled 13,885 participants, with 13,490 included in the analysis. Five studies80,126,127,131,150 involving 2426 participants used the NMP22 BladderChek point of care test. In 33 studies45,80,95,122,123,125–132,134–142,144,147–151,153,158,159,162,164 4478 participants (41%) presented with a suspicion of bladder cancer and 6536 (59%) had previously diagnosed bladder cancer. In five126,135,141,153,162 of these studies the whole patient population analysed (n = 2202) had a suspicion of bladder cancer and in 10123,127,129–131,138,142,144,147,149 the whole population analysed had previously diagnosed bladder cancer (n = 4799). Eight studies120,121,154–157,161,163 did not report this information.

Across 24 studies80,123,125–129,131,134,136,138,140,141,144,147,149–151,153,154,156,158,159,162 providing information on patient age for the whole study population, the median (range) of means was 66 years (53 to 71 years). A total of 29 studies80,121–123,125–127,129–131,135–142,144,147,149–151,153,156,158,159,162,163 provided information on the gender of 10,804 participants, of whom 7818 (72%) were men and 2986 (28%) were women.

In total, 16 studies80,121,122,126,127,135,136,139,150,151,153,155,158,159,162,164 gave details of when they took place, with an earliest start date of August 1995135 and latest end date of April 2006. 150 Nine studies took place in the USA,126,127,129,139,148,149,153,158,159 four in Italy122,123,144,154 and Spain,128,142,161,162 three in Austria,134,147,151 Germany80,95,132 and Japan,135,136,164 two in the UK,45,141 Turkey137,163 and India131,150 and one in Greece,125 Poland,130 Switzerland,121 Sweden,155 the Netherlands,138 South Korea157 and China,156 and two had multinational settings, taking place in Germany/USA140 and Saudi Arabia/USA. 120

Methodological quality of the included studies

Figure 21 summarises the quality assessment for the 41 NMP22 studies. The results for the quality assessment of the individual studies are shown in Appendix 13. In all studies the reference standard (cystoscopy with histological assessment of biopsied tissue) was considered likely to correctly classify bladder cancer (Q2) and withdrawals from the study were explained or there were none (Q11). In 40 studies (98%) the spectrum of patients who received the tests was considered to be representative of those who would receive the test in practice (Q1) and incorporation bias was avoided (Q6). In 39 studies (95%) partial verification bias was avoided (Q4), intermediate test results were reported or there were none (Q10) and a clear definition of what was considered to be a positive result was provided (Q13).

FIGURE 21.

Summary of quality assessment of NMP22 studies (n = 41).

In 36 studies (88%) differential verification bias was avoided (Q5), in 32 studies (78%) the time period between NMP22 and the reference standard was considered to be short enough (1 month or less) to be reasonably sure that the patient’s condition had not changed in the intervening period (Q3) and in 24 studies (58%) a prespecified cut-off value was used (Q12).

However, in 39 studies (95%) it was unclear whether clinical review bias had been avoided (Q9), in 29 studies (71%) it was unclear whether test review bias had been avoided (Q7) and in 27 studies (66%) it was unclear whether diagnostic review bias had been avoided (Q8). A total of 40 studies (98%) did not report information on observer variation in interpretation of test results (Q14).

Assessment of diagnostic accuracy

Patient-level analysis

A total of 28 studies45,80,95,121–123,126–128,130–132,134,137,139–142,144,147,148,150,151,153,159,160,162,163 enrolling 10,565 participants, with 10,119 included in the analysis, provided sufficient information to allow their inclusion in the pooled estimates for patient-level analysis, using a ‘common’ cut-off of 10 U/ml to define a positive test result. Figure 22 shows the sensitivity and specificity of the individual studies, pooled estimates and SROC curve for NMP22 patient-based detection of bladder cancer. The pooled sensitivity and specificity (95% CI) were 68% (62% to 74%) and 79% (74% to 84%), respectively, and the DOR value (95% CI) was 8 (5 to 11). Across the 28 studies the sensitivity for NMP22 ranged from 33%45 to 100%,153 and specificity ranged from 40%80 to 93%. 142 The median (range) PPV across studies was 52% (13% to 94%) and the median (range) NPV was 82% (44% to 100%).

FIGURE 22.

NMP22 patient-level analysis: sensitivity, specificity, pooled estimates and SROC curve.

Most of the included studies in the pooled estimates contained a mixture of patients with a suspicion of bladder cancer and those with previously diagnosed bladder cancer, although three studies121,160,163 did not report this information. In four studies126,141,153,162 all of the participants (n = 1893) had a suspicion of bladder cancer [median (range) sensitivity and specificity across studies 71% (56% to 100%) and 86% (80% to 87%) respectively]. In seven studies123,127,130,131,142,144,147 all of the participants (n = 4284) had previously diagnosed bladder cancer [median (range) sensitivity and specificity across studies 69% (50% to 85%) and 81% (46% to 93%) respectively].

Characteristics of the included studies

A description of each of the 56 included cytology studies is given in Appendix 12, which contains the characteristics of all of the included biomarker and cytology studies listed alphabetically by surname of first author. Table 18 shows summary information for the 56 cytology studies.

A total of 47 studies, reported in 56 papers, were diagnostic cross-sectional studies,45,80,97,98,100–103,105,109–129,131–133,135,136,139–141,145,146,148–153,165–174 of which 1145,101,109,110,112,116,118,126,127,165,174 reported consecutive recruitment. Nine studies107,108,155,157–159,162–164 were case–control studies and 11 studies103,108,111,116,126,127,135,149,165,170,174 were multicentre (Table 19).

The 56 studies enrolled 22,260 participants, with 19,219 included in the analysis. Eight studies80,114,120,121,151,155,167,171 involving at least 872 patients reported bladder wash cytology. In 46 studies45,80,97,101–103,105,107–114,116,118,122–124,126–129,131,132,135,136,139–141,148–151,153,158,159,162,164,165,168,170–172,174 7888 participants (45%) presented with a suspicion of bladder cancer and 9487 (55%) had previously diagnosed bladder cancer. In 10103,118,126,135,141,153,162,164,168,172 of these studies the whole patient population analysed (n = 4290) had a suspicion of bladder cancer and in 11102,108,110,111,113,123,127,129,131,170,174 the whole population analysed had previously diagnosed bladder cancer (n = 5710). In total, 10 studies98,100,119–121,155,157,163,166,167 did not report this information.

Across 33 studies80,97,101–103,107,109,112–114,116,123,124,126–129,131,136,140,141,149–151,153,158,159,162,166,170–172,174 providing information on patient age for the whole study population, the median (range) of means was 67 years (54 to 73 years). A total of 36 studies provided information on the gender of 13,341 participants, of whom 9702 (73%) were men and 3639 (27%) were women.

In total, 30 studies80,102,103,108,111–114,118,119,121,122,126,127,135,136,139,150,151,153,155,158,159,162,164–168,174 gave details of when they took place, with an earliest start date of 1990165 and latest end date of July 2007. 118 Fifteen studies took place in the USA,97,103,105,108,111,126,127,129,139,148,149,153,158,159,165 seven in Germany,80,98,107,118,132,168,171 four in the UK,45,141,166,172 three each in Italy113,122,123 and Japan,135,136,164 two each in Austria,112,151 Spain,128,161 Sweden114,155 and India131,150 and one each in Belgium,167 Finland,174 France,116 Greece,124 Switzerland,121 the Netherlands,102 Turkey,163 Canada119 and South Korea,157 while seven had multinational settings, with three taking place in Austria/Italy101,109,110 and the others taking place in Germany/USA,140 USA/Belgium,100 Saudi Arabia/USA120 and Austria/Germany/Italy/Spain/Sweden/Switzerland/Egypt/Japan/Canada/USA. 170

Methodological quality of the included studies

Figure 23 summarises the quality assessment for the 56 cytology studies. The results for the quality assessment of the individual studies are shown in Appendix 13. In all studies the reference standard (cystoscopy with histological assessment of biopsied tissue) was considered likely to correctly classify bladder cancer (Q2). In 55 studies (98%) the spectrum of patients who received the tests was considered to be representative of those who would receive the test in practice (Q1), incorporation bias was avoided (Q6), uninterpretable test results were reported or there were none (Q10) and withdrawals from the study were explained or there were none (Q11). In 54 studies (96%) partial verification bias was avoided (Q4) and in 49 (88%) differential verification bias was avoided (Q5). In 41 studies (73%) the time period between cytology and the reference standard was considered to be short enough (1 month or less) to be reasonably sure that the patient’s condition had not changed in the intervening period (Q3), in 40 studies (71%) a prespecified cut-off value for a positive test result was stated (Q12) and in 37 studies (66%) a clear definition of what was considered to be a positive result was provided (Q13).

FIGURE 23.

Summary of quality assessment of cytology studies (n = 56).

However, in 48 studies (86%) it was unclear whether clinical review bias had been avoided (Q9), in 40 studies (71%) it was unclear whether diagnostic review bias had been avoided (Q8) and in 31 studies (55%) it was unclear whether test review bias had been avoided (Q7). A total of 53 studies (95%) did not report information on observer variation in interpretation of test results (Q14).

Assessment of diagnostic accuracy

Patient-level analysis

A total of 36 studies45,80,97,100,101,107,109,111,112,116,118,119,122–124,126–128,131,132,135,136,139–141,148,150,151,153,157,159,164,166,170,172,174 reporting voided urine cytology, enrolling 15,161 participants with 14,260 included in the analysis, provided sufficient information to allow their inclusion in the pooled estimates for patient-level analysis. Figure 24 shows the sensitivity and specificity of the individual studies, pooled estimates and SROC curve for cytology patient-based detection of bladder cancer. The pooled sensitivity and specificity (95% CI) were 44% (38% to 51%) and 96% (94% to 98%), respectively, and the DOR value (95% CI) was 19 (11 to 27). Across the 36 studies the sensitivity for cytology ranged from 7%136 to 100%,172 and specificity ranged from 78%80 to 100%. 135 The median (range) PPV across studies was 80% (27% to 100%) and the median (range) NPV was also 80% (38% to 100%).

FIGURE 24.

Cytology patient-level analysis: sensitivity, specificity, pooled estimates and SROC curve.

Most of the included studies in the pooled estimates contained a mixture of patients with a suspicion of bladder cancer and those with previously diagnosed bladder cancer. In seven studies118,126,135,141,153,164,172 all of the participants (n = 3331) had a suspicion of bladder cancer [median (range) sensitivity and specificity across studies 44% (16% to 100%) and 99% (87% to 100%) respectively]. In six studies111,123,127,131,170,174 all of the participants (n = 4195) had previously diagnosed bladder cancer [median (range) sensitivity and specificity across studies 38% (12% to 47%) and 94% (83% to 97%) respectively]. Four studies100,119,157,166 did not report this information.

FISH versus cytology

Five97,98,100,101,107 of the studies included in the pooled estimates for FISH and for cytology directly compared the two tests. The studies enrolled 1377 participants, with 1119 included in the analysis for FISH and 1198 for cytology. Figure 25 shows the sensitivity and specificity of the individual studies, pooled estimates and SROC curves for these five studies. The pooled estimate (95% CI) for the sensitivity of FISH was 81% (66% to 97%) compared with 54% (39 to 80%) for cytology, whereas the pooled estimate (95% CI) for the specificity of FISH was 82% (68% to 97%) compared with 92% (84% to 99%) for cytology.

FIGURE 25.

FISH vs cytology – patient-level analysis: sensitivity, specificity, pooled estimates and SROC curve.

ImmunoCyt versus cytology

Six101,109,111,112,116,118 of the studies included in the pooled estimates for ImmunoCyt and for cytology directly compared the two tests. The studies enrolled 2016 participants, with 1912 included in the analysis. Figure 26 shows the sensitivity and specificity for the individual studies, pooled estimates and SROC curves for these six studies. The pooled estimate (95% CI) for the sensitivity of ImmunoCyt was 82% (76% to 89%) compared with 44% (35% to 54%) for cytology, whereas the pooled estimate (95% CI) for the specificity of ImmunoCyt was 85% (71% to 85%) compared with 94% (91% to 97%) for cytology.

FIGURE 26.

ImmunoCyt vs cytology – patient-level analysis: sensitivity, specificity, pooled estimates and SROC curve.

NMP22 versus cytology

In total, 1645,80,123,126–128,131,132,139–141,148,150,151,153,159 of the studies included in the pooled estimates for NMP22 and for cytology directly compared the two tests. The studies enrolled 5623 participants, with 5563 included in the analysis for NMP22 and 5402 for cytology. Figure 27 shows the sensitivity and specificity for the individual studies, pooled estimates and SROC curves for these 16 studies. The pooled estimate (95% CI) for the sensitivity of NMP22 was 70% (59% to 80%) compared with 40% (31% to 49%) for cytology, whereas the pooled estimate (95% CI) for the specificity of NMP22 was 81% (74% to 88%) compared with 97% (95% to 99%) for cytology.

FIGURE 27.

NMP22 vs cytology – patient-level analysis: sensitivity, specificity, pooled estimates and SROC curve.

FISH and cytology

Two studies94,102 reported the sensitivity and specificity of FISH and cytology used in combination. In a patient-level analysis (n = 115), Daniely and colleagues94 reported sensitivity and specificity of 100% and 50%, respectively, for FISH and cytology used in combination (results were not presented separately for the individual tests). A test was reported as positive if at least one cell abnormality in both cytology and FISH was found. In the case of abnormal FISH and normal cytology, a minimum of four cells with a gain of two or more chromosomes or 12 or more cells with homozygous loss of the 9p21 locus was required for a positive diagnosis. The study by Moonen and colleagues102 involving 105 patients reported a specimen-based analysis (n = 103), with sensitivity and specificity of 39% and 90%, respectively, for FISH, 41% and 90% for cytology and 53% and 79% for the tests used in combination.

FISH and cystoscopy

In a patient-based analysis, Kipp and colleagues99 in a study involving 124 patients reported the sensitivity and specificity of FISH and cystoscopy (not stated whether flexible or rigid) used in combination. They reported sensitivity and specificity of 62% and 87%, respectively, for FISH, 67% and 85% for cystoscopy and 87% and 79% for the tests used in combination. A definition of what constituted a positive test result for the combined tests was not given.

ImmunoCyt and cytology

Eight studies reported sensitivity and specificity for the tests of ImmunoCyt and cytology used in combination. Six studies101,109,111,112,116,119 involving 1997 patients reported patient-based detection and two studies110,113 involving 1137 patients reported specimen-based detection (2220 specimens). The median (range) sensitivities and specificities of ImmunoCyt, cytology, and ImmunoCyt and cytology across the studies reporting patient- and specimen-based detection are shown in Table 21. The sensitivity of the tests in combination for both patient- and specimen-based detection (87% and 88% respectively) was slightly higher than that of ImmunoCyt alone (84% and 78%), whereas the specificity (68% and 76%) was much lower than that of cytology alone (94% and 97%).

ImmunoCyt and cystoscopy

In a patient-based analysis (n = 280), Schmitz-Drager and colleagues118 reported sensitivity and specificity of 85% and 88%, respectively, for ImmunoCyt, 84% and 98% for cystoscopy (not stated whether flexible or rigid) and 100% and 87% for the tests used in combination.

NMP22 and cytology

In a patient-based analysis, two studies131,164 involving 800 patients reported the sensitivity and specificity of NMP22 and cytology used in combination. The study by Kumar and colleagues131 involving 131 patients used the NMP22 BladderChek point of care test with a cut-off of 10 U/ml. They reported sensitivity and specificity of 85% and 78%, respectively, for NMP22, 41% and 96% for cytology and 91% sensitivity for the tests used in combination (specificity was not reported). 131 The study by Takeuchi and colleagues164 involving 669 patients used a cut-off of 12 U/ml for NMP22. They reported sensitivity and specificity of 58% and 80%, respectively, for NMP22, 44% and 100% for cytology and 60% sensitivity for the tests used in combination (specificity was not reported). In both studies the sensitivity for the tests in combination was slightly higher than that for NMP22 alone, although there was a wide difference in the sensitivity values for NMP22 reported by the two studies.

NMP22 and cystoscopy

In a patient-based analysis (n = 1999), two studies by Grossman and colleagues126,127 reported the sensitivity and specificity of the NMP22 BladderChek point of care test and cystoscopy (not stated whether flexible or rigid) used in combination. Both studies used a cut-off of 10 U/ml to define a positive NMP22 test result. In the first study126 sensitivity was 56% and specificity 86% for NMP22 (1331 patients), whereas in 79 patients diagnosed with bladder cancer the sensitivity of cystoscopy and the tests used in combination was 89% and 94% respectively. In the second study127 sensitivity was 50% and specificity 87% for NMP22 (668 patients), whereas in 103 patients diagnosed with bladder cancer the sensitivity of cystoscopy and the tests used in combination was 91% and 99% respectively.

Cytology and cystoscopy

In a patient-based analysis (n = 280), Schmitz-Drager and colleagues118 reported sensitivity and specificity of 44% and 96%, respectively, for cytology, 84% and 98% for cystoscopy (not stated whether flexible or rigid) and 88% and 95% for the tests used in combination. In 103 patients diagnosed with bladder cancer Grossman and colleagues127 reported sensitivity of 12% for cytology, 91% for cystoscopy and 94% for the tests used in combination.

Model structure

Based on the care pathway described in Chapter 2, the model structure was developed following consultation with clinicians and taking into consideration the approaches adopted by the existing economic evaluations153,158,175–180 identified from the literature. The approach attempts to model patients passing through the whole sequence of care and determine the overall impact on costs and the clinical consequences. Figure 28 shows a simplified model structure for the primary diagnosis and follow-up management of bladder cancer. Within this model people with suspected bladder cancer will receive tests and investigations to diagnose bladder cancer. Subsequent management will depend upon the findings of these tests and the nature of any bladder cancer detected. The absorbing state in the model is death from either bladder cancer or other causes.

FIGURE 28.

Model structure.

The cost-effectiveness analysis was performed in two parts. The first part considered the diagnostic tests and consisted of a decision tree model element and the second part considered the follow-up of patients after diagnosis using a Markov model.

Interventions of diagnosis and follow-up

The interventions included in the model were flexible cystoscopy, cytology, three types of biomarkers (NMP22, FISH, ImmunoCyt), WLC and PDD. Although flexible cystoscopy combined with cytology and a biomarker as the first suite of tests may be an option for the primary diagnosis of bladder cancer, there is little information about the results of these tests used in combination, as reported in Chapters 4 and 5. Table 25 summarises the potential strategies that are considered in the model. These options were based on advice from clinical experts about strategies that are currently in use or those that can potentially be used.

TABLE 25 - Diagnostic strategies

BM, biomarker; CSC, flexible cystoscopy; CTL, cytology.

Strategy	Primary diagnosis					Follow-up surveillance
	Initial test			Second test		Initial test			Second test
	CSC	CTL	BM	WLC	PDD	CSC	CTL	BM	WLC	PDD
1	✓			✓		✓			✓
2	✓				✓	✓			✓
3		✓		✓			✓		✓
4		✓			✓		✓		✓
5			✓	✓				✓	✓
6			✓		✓			✓	✓
7	✓	✓		✓		✓			✓
8	✓	✓			✓	✓			✓
9	✓	✓		✓			✓		✓
10	✓	✓			✓		✓		✓
11	✓		✓	✓		✓			✓
12	✓		✓		✓	✓			✓
13	✓		✓	✓				✓	✓
14	✓		✓		✓			✓	✓
15	✓	✓	✓	✓		✓
16	✓	✓	✓		✓	✓

Strategies 1–6 consider the use of a single test for initial diagnosis. These options might represent situations that clinical practice might move towards although they may not be currently used in practice. Strategies 7–16 represent alternative situations in which two or more tests are used in the initial phase of diagnosis. Across all strategies the choice of second level diagnostic test varies between WLC and PDD. The strategies also differ in terms of the tests used for follow-up surveillance. In our study we have assumed that a single test is used for initial surveillance with any positives confirmed by WLC.

It should be noted that none of our strategies explicitly considers the use of ultrasound. Ultrasound might be considered part of all of the strategies when the patient population is restricted to haematuria. In such a situation this would have no impact on incremental costs (as all patients under all strategies incur the test) although it may alter the likelihood of subsequent testing.

Figure 29 illustrates a simplified model structure for the decision tree model for diagnosis of bladder cancer when a single test is used as part of the initial diagnosis (i.e. strategies 1–6). Figure 30 illustrates the model structure for the situation in which two tests are used as part of the initial testing (strategies 7–14). When three tests are used in combination (strategies 15 and 16) a similar model structure to that in Figure 30 is developed (figure not shown).

FIGURE 29.

Decision tree model structure for single diagnostic technology as the first test. BM, biomarker; CSC, flexible cystoscopy; CTL, cytology.

FIGURE 30.

Decision tree model structure for flexible cystoscopy combined with cytology or biomarker as the first test. BM, biomarker; CSC, flexible cystoscopy; CTL, cytology.

In Figure 29 a patient may, for example, arrive in a hospital with symptoms of haematuria. Taking the patient’s history and symptoms into account, the physician may perform an invasive test (flexible cystoscopy) or a non-invasive test (e.g. cytology and biomarkers). The results of these tests could be either negative or positive. The negative test result could be either a false or a true negative. If the first single test in Figure 29 is negative, it is assumed that there appears to be no evidence of bladder cancer and the patient is deemed not to have bladder cancer. If the result of the first test is positive (which might be either a true or a false positive) the patient will be further investigated using the second test, which will be either PDD or WLC. As with the first test there are four potential test results: true negative, false negative, true positive and false positive. As there is a risk of death associated with the use of general anaesthesia required for rigid cystoscopy, there is a chance that the patient may die whilst undergoing or as a result of undergoing the second test.

For the strategies in which two tests form part of the initial diagnosis (strategies 7–14) the first test that a patient receives will be flexible cystoscopy (Figure 30). If the result is negative (it might be either a true or a false negative) it is assumed that the patient will be further tested using cytology or a biomarker. If the result of cytology or a biomarker is negative the patient will be deemed not to have bladder cancer. If the result of the first test is positive (which might be either a true or a false positive) the patient will be further investigated using the second test, which will be either PDD or WLC. Patients who test positive with cytology or a biomarker will be handled in a similar manner. As with the first test there are four potential test results: true negative, false negative, true positive and false positive. As there is a risk of death associated with the use of general anaesthesia required for rigid cystoscopy, there is a chance that the patient may die whilst undergoing or as a result of undergoing the second test.

Strictly speaking, Figure 30 describes the situation in which only those negative on flexible cystoscopy (CSC) receive either cytology (CTL) or a biomarker (BM) test. In practice, because of the way that services might be organised, the different tests may be performed during the same visit, i.e. those who are positive with flexible cystoscopy may also receive either cytology or a biomarker test. The implications of this are that, given the cost data available for this study, the average cost per patient in actual practice would be increased compared with the practice described in Figure 29 (there will be no impact on effectiveness as all positives go through to the next level of testing). It should be noted that the practice of conducting additional tests at the same time as flexible cystoscopy is likely to be adopted because it is logistically easier to organise, i.e. the real opportunity costs of current practice are less than would be predicted from the unit costs available for this study. For this reason we have assumed that a more realistic estimate of costs will be provided by a model following the structure set out in Figure 30 but we have provided an additional analysis to illustrate the effect on costs when two or more tests are conducted on all patients presenting for initial diagnosis.

Estimation of probabilities required for the decision tree model

The probabilities used to populate the decision model were calculated according to the standard conventions of Bayes’ theorem. The essence of the calculations is that, once the sensitivity and specificity of a test are known, along with the a priori probability of disease, the posterior probabilities of disease and absence of disease can be determined. Accordingly, if a patient has an abnormal test result, the probability of disease – the ‘true positive rate’, also referred to as the ‘positive predictive value’ (PPV) – is represented as p(BC+|T+), and if the patient has a normal test result, the probability of disease – the ‘false-negative rate’ – is similarly presented as p(BC+|T–). These are calculated as follows:

p ( BC + | T + ) = p ( T + | BC + ) p ( BC + ) / ( p ( T + | BC + ) p ( BC + ) + p ( T + | BC – ) p ( BC – ) )

p ( BC + | T – ) = p ( T – | BC + ) p ( BC + ) / ( p ( T – | BC + ) p ( BC + ) + p ( T – | BC – ) p ( BC – ) )

where BC = bladder cancer, T+ = test positive, T– = test negative, p(T+|BC+) = sensitivity, p(BC+) = prior probability of disease (prevalence or incidence), p(T+|BC–) = 1 – specificity, p(BC–) = 1 – prevalence (or incidence), p(T–|BC+) = 1 – sensitivity and p(T–|BC–) = specificity.

When two tests are connected in series, the calculations are the same except that the prior probability of disease (prevalence or incidence) for the second test is the calculated ‘true positive rate’ of the first test.

To illustrate this in the construction and analysis of the bladder cancer primary diagnosis tree (Appendix 17, Figure 37), the strategy ‘flexible cystoscopy (CSC) followed by WLC’ is considered. The probability of a test positive result following flexible cystoscopy is:

pPos _ CSC = ( Sens _ CSC * priori ) + ( 1 – Spec _ CSC ) * ( 1 – priori )

where Sens_CSC = sensitivity of flexible cystoscopy, Spec_CSC = specificity of flexible cystoscopy and priori is the prevalence or incidence rate for patients with suspected bladder cancer before the flexible cystoscopy test.

From this, the probability of a:

• negative result for flexible cystoscopy is 1 – pPos_CSC

• false negative for flexible cystoscopy is pFN_CSC = (1 – Sens_CSC)*priori/((1 – Sens_CSC)* priori + Spec_CSC*(1 – priori))

• true negative is 1 – pFN_CSC

• positive result for WLC following a positive flexible cystoscopy result is pPos_CSC_WLC = (Sens_WLC*pPPV_CSC) + (1 – Spec_WLC)*(1 – pPPV_CSC), where Sens_WLC = sensitivity of WLC, Spec_WLC = specificity of WLC and pPPV_CSC = positive predictive value of flexible cystoscopy = (Sens_CSC*priori)/pPos_CSC

• true positive for WLC following a positive flexible cystoscopy result is pTP_CSC_WLC = (Sens_WLC*pPPV_CSC)/pPos_CSC_WLC

• false positive for WLC following flexible cystoscopy is 1 – pTP_CSC_WLC

• false negative for WLC following flexible cystoscopy is pFN_CSC_WLC = [Spec_WLC*(1 – pPPV_CSC)]/(1 – pPos_CSC_WLC)

• true negative is 1 – pFN_CSC_WLC

• the NPV after a negative result for CSC is pNPV_CSC = [Spec_CSC*(1 – priori)]/(1 – pPos_CSC).

The probabilities for the remaining strategies in the tree are calculated in a similar manner.

It is important to quantify the false-positive and false-negative values for each strategy, as these provide valuable information to the clinician in addition to the cost and number of true cases detected. The implications of false-positive results within the model are the cost of testing and treating patients and the associated morbidity and discomfort of further investigation and treatment. False-positive results may also induce adverse psychological responses in patients in terms of the needless distress that a positive result might cause and by leading to questioning of future results that are negative. In the case of false-negative results the patient may have a serious or life-threatening condition that is missed, resulting in a potentially poorer prognosis following late detection, such as CIS missed by WLC, as well as psychological distress from false reassurance. In the decision model patients with a false-negative evaluation following the first (flexible cystoscopy, cytology or biomarkers) or second (PDD/WLC) test may be subsequently correctly diagnosed as their continuing symptoms worsen. In the case of true negative results, it is assumed that the patients will not need further investigation.

Management of bladder cancer

Patients with true-positive results (confirmed bladder cancer) are classified into two types: non-muscle-invasive and muscle-invasive disease (Figure 31). Those with muscle-invasive tumours will not be discharged but are managed usually with either surgery (radical cystectomy) or radical radiotherapy with or without chemotherapy and routine checking thereafter and treatment. All patients with non-muscle-invasive tumours will undergo a follow-up test at 3 months after the primary diagnosis because of the high chance of recurrence and a chance of progression. For each risk group there are similar outcomes considered in initial diagnosis: true positive, false positive, true negative and false negative (Appendix 17, Figure 37).

FIGURE 31.

Classification of bladder cancer.

It is assumed that the first test used in the follow-up of patients will be the same as the test used for primary diagnosis and the second test will be WLC. To illustrate the construction and analysis of each risk group, strategy ‘flexible cystoscopy (CSC) followed by WLC in primary diagnosis and follow-up by CSC’ is considered. In the case of each group, the probability of:

• true positive is pTP_Riskgroup = Sens_CSC* Recurrence rate of risk group at 3 months

• true negative is pTN_Riskgroup = Spec_CSC*(1 – Recurrence rate of risk group at 3 months)

• false negative is pTP_Riskgroup = (1 – Sens_CSC)* Recurrence rate of risk group at 3 months

• false positive is pFN_Riskgroup = (1 – Spec_CSC)*(1 – Recurrence rate of risk group at 3 months).

As described in the care pathway reported in Chapter 2, bladder cancer treatment options will depend on classification of disease (Table 26).

TABLE 26 - Management of bladder cancer

Type of bladder cancer		Initial treatment
Non-muscle invasive	Low risk	TURBT and one dose of mitomycin
	Intermediate risk	TURBT and one dose of mitomycin
	High risk	TURBT, one dose of mitomycin and BCG induction
Muscle invasive		Three cycles of chemotherapy and cystectomy or three cycles of chemotherapy and radiotherapy or palliative treatment

To determine the efficiency of each strategy the terminal nodes (Appendix 17, Figure 37) of the tree were assigned a value of either ‘1’ or ‘0’. This enabled the following solutions to be calculated: mean cost per case detected – achieved by assigning the value ‘0’ to dead terminal node and the value ‘1’ to the others.

Markov model structure for non-muscle-invasive disease

At the end of each risk group branch of non-muscle-invasive disease in the decision tree (Appendix 17, Figure 36) the patient will enter one of the following states of the Markov model shown in Figure 32: (1) no tumour recurrence; (2) recurrence; (3) progression to muscle-invasive disease; and (4) death. There are two diagnostic results of non-tumour recurrence, i.e. true negative and false negative, as well as true positive and false positive for tumour recurrence.

FIGURE 32.

Markov model structure for non-muscle-invasive tumour.

The patients with a false-negative result in the model will be followed using the follow-up strategy of non-tumour recurrence. The cycle length considered is 1 year, although the risk groups in the care pathway will be followed at different time periods: 12 months for low risk, 6 months for intermediate risk and 3 months for high risk. The absorbing state is ‘death’, which can be reached from any of the other states.

Markov model for local muscle-invasive disease

At the end of each risk group branch of local muscle-invasive disease in the decision tree (Appendix 17, Figure 38) the patient will enter one of the following states of the Markov model shown in Figure 33: (1) no tumour; (2) recurrence; (3) progression to metastases; and (4) death. Cycle length will be the same as that of non-muscle-invasive disease.

FIGURE 33.

Markov model for local muscle-invasive follow-up.

Estimation of parameters used in the model

Parameters used in the decision tree and Markov models were calculated within the model or estimated from the systematic reviews of diagnostic performance reported in Chapters 4 and 5 and the epidemiology of bladder cancer reported in Chapter 1, as well as other relevant cost-effectiveness data identified from the literature. The details of the data for the probabilities, costs and utilities used in the models are described below.

Probabilities

Sensitivity and specificity of diagnostic test

The data on the sensitivity and specificity of each diagnostic test were taken from the systematic review and are summarised in Table 28. For flexible cystoscopy assessment there were no data available from the systematic review. It is therefore assumed that the accuracy of flexible cystoscopy used in the models is the same as that of white light rigid cystoscopy. This assumption is relaxed in the sensitivity analysis in which the performance of flexible cystoscopy is increased by 5%, 10% and an extreme 20% compared with white light rigid cystoscopy.

TABLE 28 - Data on diagnostic performance

CSC, flexible cystoscopy; CTL, cytology.

Diagnosis	Sensitivity	95% CI	Specificity	95% CI	Source
CSC	0.71	0.49 to 0.93	0.72	0.47 to 0.96	Systematic review based on WLC
CTL	0.44	0.38 to 0.51	0.96	0.94 to 0.98	Systematic review
NMP22	0.68	0.62 to 0.74	0.79	0.74 to 0.84	Systematic review
ImmunoCyt	0.84	0.79 to 0.91	0.75	0.68 to 0.83	Systematic review
FISH	0.76	0.65 to 0.84	0.85	0.78 to 0.92	Systematic review
PDD	0.92	0.8 to 1.0	0.57	0.36 to 0.79	Systematic review
WLC	0.71	0.49 to 0.93	0.72	0.47 to 0.96	Systematic review

Prevalence rate

The prevalence rate was not derived from existing data in the literature as the prevalence of bladder cancer varies considerably among subgroups with different symptoms, from 1% to 20% (for men over 50 years of age). 179 In the model base-case analysis it was assumed that the prevalence rate is 5% and in a sensitivity analysis a range of prevalence rates was considered to identify those prevalence rates for which different diagnostic strategies may be considered worthwhile. This approach of repeating the analysis for different prevalence rates was felt to be more informative than defining prevalence using a wide uniform (i.e. uninformative) distribution.

Proportions of types and their subgroups for bladder cancer

The proportions of the two main types of bladder cancer were assessed based on the literature and clinical opinions detailed in Chapter 1. With reference to the available information presented in the previous section and in Table 27, as well as discussions with the clinical members of the research team, prognostic risk groups in non-muscle-invasive disease within this model have been categorised by using a combination of Millán-Rodriguez and colleagues’187 classification at initial diagnosis and Parmar and colleagues’191 classifications at 3 months’ follow-up, i.e. low risk, intermediate risk and high risk. These classifications are shown in Table 29, which also provides details on the proportions of patients in each risk group of non-muscle-invasive bladder cancer.

TABLE 29 - Risk group stratification

If TaG3, T1G3, CIS, multi T1G2 recurrence, then joins high-risk treatment pathway.

Risk groups	Subgroups (cancer at diagnosis)	Factors defined in follow-up at 3 months	Proportion (%)
Low: TaG1, single T1G1	Group 1: single TaG1, single T1G1	No tumour recurrence	10
	Group 2a: single TaG1, single T1G1	Tumour recurrencea
	Group 2b: multi TaG1	No tumour recurrence
	Group 3: multi TaG1	Tumour recurrencea
Intermediate: TaG2, multi T1G1, single T1G2	Group 1: single TaG2, single T1G2	No tumour recurrence	45
	Group 2a: single TaG2, single T1G2	Tumour recurrencea
	Group 2b: multi TaG2, multi T1G1	No tumour recurrence
	Group 3: multi TaG2, multi T1G1	Tumour recurrencea
High: TaG3, T1G3, CIS, multi T1G2		Tumour recurrence or not	45

Table 30 summarises the values of these proportions used in the decision tree and Markov models.

TABLE 30 - Proportions of types and their subgroups for bladder cancer

Type of bladder cancer	Proportion	Subgroups of bladder cancer considered	Proportion
Non-muscle invasive	75%	Low risk	10%
		Intermediate risk	45%
		High risk	45%
Muscle invasive	25%	Local muscle invasive	75%
		Metastases	25%

Recurrence, progression and mortality of non-muscle-invasive disease

Table 31 shows the probabilities of recurrence, progression and mortality for the three risk groups of non-muscle-invasive disease used in the model for a 20-year time horizon. As referred to above, the first 5-year probabilities of recurrence, progression and mortality caused by cancer of the three risk groups used in the model were calculated from the study by Millán-Rodriguez and colleagues. 187 The following 15-year probabilities of recurrence, progression and mortality caused by cancer in these groups were estimated by using mean values of relevant data of the last 3 years in the 5-year data available in the study by Millán-Rodriguez and colleagues. 187 This was a retrospective cohort study of 1529 patients with primary non-muscle-invasive bladder cancer in Spain in the years 1968–96. Of the patients treated with TURBT and random biopsy, half were treated using additional BCG and one-third using additional intravesical instillation (mainly mitomycin C, thiotepa and doxorubicin). However, the characteristics of the patients, such as gender and mean age, were not reported, and the follow-up was less than 5 years. 187

TABLE 31 - Probabilities of recurrence, progression and mortality in non-muscle-invasive bladder cancer

Time (years)	Recurrence (%)			Progression (%)			Mortality caused by cancer (%)
	Low	Intermediate	High	Low	Intermediate	High	Low	Intermediate	High
3 months	2	4	9.4	0	0.2	1.3	0	0	0
1	15	26	39	0	0.4	8	0	0.4	1
2	10	13	11	0	0.8	5	0	0	3
3	5	6	6	0	0.6	3	0	0.3	1
4	8	5	2	0	0.8	1	0	0.3	2
5	7	3	3	0	1	2	0	0	2
6	7	5	4	0	0.8	2	0	0.2	2
7	7	5	4	0	0.8	2	0	0.2	2
8	7	5	4	0	0.8	2	0	0.2	2
9	7	5	4	0	0.8	2	0	0.2	2
10	7	5	4	0	0.8	2	0	0.2	2
11	7	5	4	0	0.8	2	0	0.2	2
12	7	5	4	0	0.8	2	0	0.2	2
13	7	5	4	0	0.8	2	0	0.2	2
14	7	5	4	0	0.8	2	0	0.2	2
15	7	5	4	0	0.8	2	0	0.2	2
16	7	5	4	0	0.8	2	0	0.2	2
17	7	5	4	0	0.8	2	0	0.2	2
18	7	5	4	0	0.8	2	0	0.2	2
19	7	5	4	0	0.8	2	0	0.2	2
20	7	3	4	0	0.8	2	0	0.2	2

Recurrence, progression and mortality of muscle-invasive disease

When patients move into the Markov model for muscle-invasive disease, the model requires estimates of the annual rates of recurrence, progression and mortality caused by cancer. The probabilities of recurrence, progression and mortality of muscle-invasive disease and metastases used in the model for 20 years are presented in Table 32. The first 5-year probabilities of recurrence, progression and mortality caused by local muscle-invasive disease used in the model were obtained from a retrospective cohort study in Canada by Stein and colleagues197 in which a cohort of 1054 patients with muscle-invasive bladder cancer were treated by radical cystectomy between 1971 and 1997. The mean age of the patients was 66 years, 80% of the patients were male197 and data were available for 10 years of follow-up. The last 10-year probabilities used in the model are assumed to be the same as the data reported for between 5 and 10 years in the study by Stein and colleagues. The last column of Table 32 presents the probabilities of mortality for metastases provided by von der Maase and colleagues198 and there are data available for 5 years of follow-up. The last 5-year probabilities used in the model are assumed to be the same as rates reported for between 3 and 5 years in von der Maase and colleagues. 198 This RCT investigated the long-term survival of patients with metastatic bladder cancer treated with chemotherapy in Denmark. Of the 405 patients, 137 had locally advanced disease and 268 had metastatic disease. The median survival time was 8.3 months.

TABLE 32 - Probabilities of recurrence, progression and mortality in muscle-invasive bladder cancer

Time (years)	Local muscle-invasive disease after cystectomy			Metastases
	Recurrence (%)	Progression (%)	Mortality (%)	Mortality (%)
3 months	0	6.25	3	10.5
1	0	25	12	42
2	0	13	11	80
3	0	8	9	50
4	0	4	8	50
5	0	4	8	50
6	0	4	7	50
7	0	4	6	50
8	0	4	5	50
9	0	4	5	50
10	0	4	5	50
11	0	4	5
12	0	4	5
13	0	4	5
14	0	4	5
15	0	4	5
16	0	4	5
17	0	4	5
18	0	4	5
19	0	4	5
20	0	4	5

All-cause mwortality rates in the UK

As patients progress through the model over time, values of annual rates of age-specific general or all-cause mortality are required. These were taken from the published UK life tables for the years 2004–6. 199 As discussed in Chapter 1, Cancer Research UK reported that 70% of all primary bladder cancer affects men and therefore the all-cause mortality for the model cohort was weighted to reflect this (Figure 34). Further data related to the rate of all-cause mortality are shown in Table 57 in Appendix 18.

FIGURE 34.

Kaplan–Meier plot for sex- and age-adjusted survival (30% female, 70% male) in the UK.

Other probabilities

Mortality rates of WLC/PDD and TURBT

White light rigid cystoscopy (WLC), PDD and TURBT are invasive procedures. As with all surgical procedures requiring general anaesthetic, death due to complications in the perioperative period is a potential risk. There are no available data on mortality rates associated with WLC or PDD. The probability of death during WLC and PDD in Table 33 was therefore obtained from a study by Farrow and collegues,200 which examined 108,878 anaesthetic cases in Cardiff between 1972 and 1977. The probability of death during TURBT in Table 33 was obtained from Kondas and colleagues,201 which evaluated 1250 TURBT cases in Cardiff during 18 years.

TABLE 33 - Other probabilities

Other probabilities	Value	Source
Mortality rate of WLC/PDD	0.5%	Farrow 1982200
Mortality rate of TURBT	0.8%	Kondas 1992201
False negatives: probability detected in first 3 months	50%	Assumption
Relative risk for progression (no treatment vs treatment)	2.56	Millán-Rodriguez 2000187
Relative risk for recurrence (PDD vs WLC)	1	Assumption
Relative risk for progression (PDD vs WLC)	1	Assumption
False negatives: probability detected in first year	50%	Assumption
False negatives: probability detected in second year	75%	Assumption
False negatives: probability detected after second year	100%	Assumption

Relative risk for progression comparing no treatment (false negative) with treatment (true positive)

As some patients who have bladder cancer show negative results during the initial diagnosis or follow-up, it was believed that the risk of progression in the case of a false negative without relevant treatment was higher than that of a true positive with treatment. However, there are no data available in relation to false-negative diagnoses. Although there are some studies investigating disease-free survival or survival for different types of drug treatment as an adjunct to initial treatment (TURBT) for bladder cancer, there is no identified study that compares survival with and without TURBT. Using information from the Millán-Rodriguez and colleagues’ study187 it was assumed that the base-case RR for progression comparing no treatment (TURBT) with treatment (TURBT) was 2.56, that is the RR compared TURBT plus BCG with TURBT alone. The uncertainty around this value was tested as part of the sensitivity analysis.

Relative risks for recurrence and progression comparing PDD with WLC treatment

One of the issues that could be considered in the model is whether the recurrence and progression rates of non-muscle-invasive disease differ based on the type of intervention used in the treatment (PDD or WLC). Although there is some evidence in Chapter 4 that PDD may reduce recurrence and progression for non-muscle-invasive disease compared with WLC, there are no reliable data related to recurrence and progression of non-muscle-invasive bladder cancer following PDD or WLC in primary diagnosis. It was therefore assumed that recurrence and progression rates are not different between PDD and WLC so that the base-case RR for recurrence and progression comparing PDD and WLC is 1. This assumption was tested as part of the sensitivity analysis.

Probability of detecting missed bladder cancer after false-negative results

There is no evidence to suggest when patients who have false-negative results should be detected. Therefore assumptions were made about when such patients were identified. The probabilities of detecting false-negative cases are described in Table 33.

Data analysis

Deterministic and probabilistic results

The cost-effectiveness analysis aggregates the diagnostic performance and the time spent in the various health states of the model. As described previously cost–utility analysis was not performed because QoL data suitable for incorporation into the economic model were not available.

Probabilistic results

The cost-effectiveness point estimates do not provide any information on uncertainty surrounding the model parameters. The results of the probabilistic analysis revealed the level of uncertainty concerning results as illustrated in the CEACs in Figure 35.

FIGURE 35.

Cost-effectiveness acceptability curves determined by society’s willingness to pay for a life-year for the eight strategies.

As can be seen in Figure 35 none of the eight strategies considered is likely to be cost-effective more than 50% of the time when society is willing to pay relatively little for an additional life-year except for strategy 1 [CTL_WLC (CTL-WLC)]. Nevertheless, there are four strategies that are each associated with an approximately 20% chance of being considered cost-effective over much of the range of willingness to pay values considered. It is notable that three of the four strategies involve the use of biomarkers for diagnosis and follow-up, while the fourth uses cytology.

As mentioned in the methods section of this chapter, the cost-effectiveness estimates for those strategies that involve more than one test as part of the initial diagnosis may be underestimated. Adding in these potential extra costs had virtually no effect on the point estimates of cost-effectiveness or on the likelihood that a particular strategy would be likely to be considered cost-effective.

Sensitivity analysis and subgroup analysis