Katie Gustafson

Urinary biomarkers for diagnosis of bladder cancer: A systematic review and meta-analysis

Bladder cancer is the fourth most commonly diagnosed cancer in U.S. men and the 10th most commonly diagnosed cancer in U.S. women (1). Standard methods for diagnosis of bladder cancer involve cytologic evaluation of urine, imaging tests, and cystoscopy (2, 3). Because cystoscopy is uncomfortable and costly, alternative diagnostic methods have been sought. Urine-based biomarkers have been developed as potential alternatives or adjuncts to standard tests for the initial diagnosis of bladder cancer or identification of recurrent disease (4). Six urinary biomarkers have been approved by the U.S. Food and Drug Administration (FDA) for diagnosis or surveillance of bladder cancer: quantitative nuclear matrix protein 22 (NMP22) (Alere NMP22 [Alere]), qualitative NMP22 (BladderChek [Alere]), qualitative bladder tumor antigen (BTA) (BTA stat [Polymedco]), quantitative BTA (BTA TRAK [Polymedco]), fluorescence in situ hybridization (FISH) (UroVysion [Abbott Molecular]), and fluorescent immunohistochemistry (ImmunoCyt [Scimedx]). The qualitative NMP22 and BTA tests can be performed as point-of-care tests, and the others are performed in a laboratory. One additional test, Cxbladder (Pacific Edge Diagnostics USA), is a laboratory-developed test that does not require FDA approval. Other biomarkers have been developed but are not FDA-approved. The purpose of this study was to systematically review the evidence on the comparative accuracy of urinary biomarkers for diagnosis of bladder cancer. It was done as part of a larger review (5) on the evaluation and treatment of nonmuscle-invasive bladder cancer that was nominated to the Agency for Healthcare Research and Quality (AHRQ) by the American Urological Association for use in updating its guidelines. Methods Detailed methods and data for this review, including the analytic framework, key questions, search strategies, inclusion criteria, study data extraction, and quality ratings, are available in the full report (5). The protocol was developed using a standardized process (6) with input from experts and the public and is registered in the PROSPERO database (7). This article focuses on the accuracy of urinary biomarkers for initial diagnosis of bladder cancer or for diagnosis of recurrent disease, including any variance in diagnostic accuracy based on tumor characteristics, patient characteristics, or the nature of presenting signs or symptoms. Data Sources and Searches A research librarian searched multiple electronic databases, including Ovid MEDLINE (January 1990 through June 2015), the Cochrane Central Register of Controlled Trials, and the Cochrane Database of Systematic Reviews (through June 2015). We also reviewed reference lists and searched ClinicalTrials.gov. Study Selection Two investigators independently reviewed abstracts and full-text articles against prespecified eligibility criteria. We included cross-sectional and cohort studies on the diagnostic accuracy of urinary biomarkers in adults who had signs or symptoms of bladder cancer or were undergoing surveillance for recurrent disease after treatment. We focused on urinary biomarkers approved by the FDA for the diagnosis of bladder cancer (quantitative or qualitative NMP22, qualitative or quantitative BTA, FISH, and ImmunoCyt) or classified by the FDA as a laboratory-developed test (Cxbladder). We excluded studies that used a casecontrol design; studies that did not evaluate the diagnostic accuracy of biomarkers against standard diagnostic methods (cystoscopy and histopathology); and studies on the accuracy of biomarkers for screening in assessing prognosis, guiding therapy, or monitoring response to treatment. Data Extraction and Quality Assessment One investigator extracted details about the setting, tests evaluated, definition of a positive test result, study design, reference standard, inclusion criteria, population characteristics, proportion found to have bladder cancer, bladder cancer stage and grade, results, and funding sources. We constructed 22 tables with the number of true-positive, false-positive, true-negative, and false-negative results from published sample sizes, prevalence, sensitivity, and specificity. A second investigator verified extractions for accuracy. Two investigators independently assessed the risk of bias for each study as low, moderate, or high using criteria adapted from QUADAS-2 (Quality Assessment of Diagnostic Accuracy Studies 2) (8). Discrepancies were resolved through discussion and consensus. Data Synthesis and Analysis We performed meta-analyses for sensitivity and specificity using a bivariate logistic mixed-effects model (9) with SAS, version 10.0 (SAS Institute) (10). We assumed random effects with a bivariate normal distribution and measured statistical heterogeneity with the random-effects variance (2). When few studies were available for an analysis, we used the moment estimates of correlation between sensitivity and specificity in the bivariate model. We calculated positive and negative likelihood ratios (LRs) using the summarized sensitivity and specificity (11, 12). Because studies of a particular biomarker generally used the same definition for a positive test result, we did not plot summary receiver-operating characteristic curves (13). For head-to-head comparisons, we used the same bivariate logistic mixed-effects model, with an added indicator variable for the tests. We conducted analyses for each biomarker by using data from all patients and data stratified according to whether testing was performed for initial diagnosis (evaluation of symptoms) or diagnosis of recurrence (surveillance). We also performed analyses stratified by study design features (such as retrospective or prospective or use of a prespecified threshold to define a positive test result), risk of bias (overall and whether the study performed blinding to the results of the index test), the country in which the study was conducted, and tumor grade and stage (14). We assessed the strength of evidence (SOE) for each body of evidence as high, moderate, low, or insufficient based on aggregate study quality, precision, consistency, and directness. Role of the Funding Source This project was funded under contract HHSA290201200014I from the AHRQ, U.S. Department of Health and Human Services. AHRQ staff assisted in developing the scope and key questions. The AHRQ had no role in study selection, quality assessment, or synthesis. Results The literature flow diagram (Figure 1) summarizes the search and selection of articles. Database searches resulted in 4358 potentially relevant articles. After dual review of abstracts and titles, we selected 262 articles for full-text dual review and determined that 57 studies (in 60 publications) met our inclusion criteria (Appendix Table 1) (15-74). Nineteen studies evaluated quantitative NMP22, 4 evaluated qualitative NMP22, 23 evaluated qualitative BTA, 4 evaluated quantitative BTA, 10 evaluated FISH, 13 evaluated ImmunoCyt, and 1 evaluated Cxbladder. Sample sizes ranged from 26 to 3916, mean age ranged from 54 to 77 years, the proportion of male patients ranged from 57% to 88%, and the proportion diagnosed with bladder cancer ranged from 3% to 81%. Eight studies focused on diagnostic testing for signs and symptoms suggestive of bladder cancer, 16 focused on surveillance of previously treated bladder cancer, and 19 evaluated mixed populations. Forty-three studies were conducted in the United States or Europe. We rated 2 studies as having low risk of bias (20, 21), 3 as having high risk of bias (25, 62, 68), and the remainder as having medium risk of bias. Frequent methodological shortcomings were failure to report blinded interpretation of the reference standard, failure to report enrollment of a random or consecutive sample of patients, or failure to report predefined criteria for a positive test result. Figure 1. Summary of evidence search and selection. * Cochrane Central Register of Controlled Trials and Cochrane Database of Systematic Reviews. Includes prior reports, reference lists of relevant articles, and systematic reviews. Appendix Table 1. Biomarker Study Characteristics Appendix Table 1 Continued Appendix Table 1 Continued Appendix Table 1 Continued Quantitative NMP22 Sensitivity of quantitative NMP22 was 0.69 (95% CI, 0.62 to 0.75), and specificity was 0.77 (CI, 0.70 to 0.83) (19 studies), for a positive LR of 3.05 (CI, 2.28 to 4.10) and a negative LR of 0.40 (CI, 0.32 to 0.50) (Appendix Figure 1). Exclusion of 2 studies that used a cutoff other than >10 U/mL for a positive test result (18, 37) resulted in similar sensitivity and specificity. Diagnostic accuracy was similar for evaluation of symptoms and for surveillance. Excluding 1 study with high risk of bias (68) and restricting the analysis to prospective studies, those conducted in the United States or Europe, or those that used a prespecified threshold for a positive test result had little effect on pooled estimates. Restricting the analysis to 3 studies with blinded reference standard interpretation resulted in higher specificity (0.89 [CI, 0.78 to 0.95]) (15, 42, 58). Appendix Figure 1. Sensitivity and specificity of quantitative NMP22. NMP22 = nuclear matrix protein 22; TN = true-negative; TP = true-positive. Qualitative NMP22 Sensitivity of qualitative NMP22 was 0.58 (CI, 0.39 to 0.75), and specificity was 0.88 (CI, 0.78 to 0.94) (4 studies), for a positive LR of 4.89 (CI, 3.23 to 7.40) and a negative LR of 0.48 (CI, 0.33 to 0.71) (Appendix Figure 2) (20, 21, 23, 37). Restricting the analysis to 2 studies with low risk of bias resulted in similar estimates (sensitivity, 0.53 [CI, 0.29 to 0.75]; specificity, 0.87 [CI, 0.74 to 0.94]) (20, 21). Subgroup and sensitivity analyses were limited by small numbers of studies. Appendix Figure 2. Sensitivity and specificity of qualitative NMP22. NMP22 = nuclear matrix protein 22; TN = true-negative; TP = true-positive. Qualitative BTA Sensitivity of qualit

Treatment of muscle‐invasive bladder cancer: A systematic review

There is uncertainty regarding the use of bladder‐sparing alternatives to standard radical cystectomy, optimal lymph node dissection techniques, and optimal chemotherapeutic regimens. This study was conducted to systematically review the benefits and harms of bladder‐sparing therapies, lymph node dissection, and systemic chemotherapy for patients with clinically localized muscle‐invasive bladder cancer. Systematic literature searches of MEDLINE (from 1990 through October 2014), the Cochrane databases, reference lists, and the ClinicalTrials.gov Web site were performed. A total of 41 articles were selected for review. Bladder‐sparing therapies were found to be associated with worse survival compared with radical cystectomy, although the studies had serious methodological shortcomings, findings were inconsistent, and only a few studies evaluated currently recommended techniques. More extensive lymph node dissection might be more effective than less extensive dissection at improving survival and decreasing local disease recurrence, but there were methodological shortcomings and some inconsistency. Six randomized trials found cisplatin‐based combination neoadjuvant chemotherapy to be associated with a decreased mortality risk versus cystectomy alone. Four randomized trials found adjuvant chemotherapy to be associated with decreased mortality versus cystectomy alone, but none of these trials reported a statistically significant effect. There was insufficient evidence to determine optimal chemotherapeutic regimens. Cancer 2016;122:842–51.

Urinary biomarkers for diagnosis of bladder cancer

Bladder cancer is the fourth most commonly diagnosed cancer in U.S. men and the 10th most commonly diagnosed cancer in U.S. women (1). Standard methods for diagnosis of bladder cancer involve cytologic evaluation of urine, imaging tests, and cystoscopy (2, 3). Because cystoscopy is uncomfortable and costly, alternative diagnostic methods have been sought. Urine-based biomarkers have been developed as potential alternatives or adjuncts to standard tests for the initial diagnosis of bladder cancer or identification of recurrent disease (4). Six urinary biomarkers have been approved by the U.S. Food and Drug Administration (FDA) for diagnosis or surveillance of bladder cancer: quantitative nuclear matrix protein 22 (NMP22) (Alere NMP22 [Alere]), qualitative NMP22 (BladderChek [Alere]), qualitative bladder tumor antigen (BTA) (BTA stat [Polymedco]), quantitative BTA (BTA TRAK [Polymedco]), fluorescence in situ hybridization (FISH) (UroVysion [Abbott Molecular]), and fluorescent immunohistochemistry (ImmunoCyt [Scimedx]). The qualitative NMP22 and BTA tests can be performed as point-of-care tests, and the others are performed in a laboratory. One additional test, Cxbladder (Pacific Edge Diagnostics USA), is a laboratory-developed test that does not require FDA approval. Other biomarkers have been developed but are not FDA-approved. The purpose of this study was to systematically review the evidence on the comparative accuracy of urinary biomarkers for diagnosis of bladder cancer. It was done as part of a larger review (5) on the evaluation and treatment of nonmuscle-invasive bladder cancer that was nominated to the Agency for Healthcare Research and Quality (AHRQ) by the American Urological Association for use in updating its guidelines. Methods Detailed methods and data for this review, including the analytic framework, key questions, search strategies, inclusion criteria, study data extraction, and quality ratings, are available in the full report (5). The protocol was developed using a standardized process (6) with input from experts and the public and is registered in the PROSPERO database (7). This article focuses on the accuracy of urinary biomarkers for initial diagnosis of bladder cancer or for diagnosis of recurrent disease, including any variance in diagnostic accuracy based on tumor characteristics, patient characteristics, or the nature of presenting signs or symptoms. Data Sources and Searches A research librarian searched multiple electronic databases, including Ovid MEDLINE (January 1990 through June 2015), the Cochrane Central Register of Controlled Trials, and the Cochrane Database of Systematic Reviews (through June 2015). We also reviewed reference lists and searched ClinicalTrials.gov. Study Selection Two investigators independently reviewed abstracts and full-text articles against prespecified eligibility criteria. We included cross-sectional and cohort studies on the diagnostic accuracy of urinary biomarkers in adults who had signs or symptoms of bladder cancer or were undergoing surveillance for recurrent disease after treatment. We focused on urinary biomarkers approved by the FDA for the diagnosis of bladder cancer (quantitative or qualitative NMP22, qualitative or quantitative BTA, FISH, and ImmunoCyt) or classified by the FDA as a laboratory-developed test (Cxbladder). We excluded studies that used a casecontrol design; studies that did not evaluate the diagnostic accuracy of biomarkers against standard diagnostic methods (cystoscopy and histopathology); and studies on the accuracy of biomarkers for screening in assessing prognosis, guiding therapy, or monitoring response to treatment. Data Extraction and Quality Assessment One investigator extracted details about the setting, tests evaluated, definition of a positive test result, study design, reference standard, inclusion criteria, population characteristics, proportion found to have bladder cancer, bladder cancer stage and grade, results, and funding sources. We constructed 22 tables with the number of true-positive, false-positive, true-negative, and false-negative results from published sample sizes, prevalence, sensitivity, and specificity. A second investigator verified extractions for accuracy. Two investigators independently assessed the risk of bias for each study as low, moderate, or high using criteria adapted from QUADAS-2 (Quality Assessment of Diagnostic Accuracy Studies 2) (8). Discrepancies were resolved through discussion and consensus. Data Synthesis and Analysis We performed meta-analyses for sensitivity and specificity using a bivariate logistic mixed-effects model (9) with SAS, version 10.0 (SAS Institute) (10). We assumed random effects with a bivariate normal distribution and measured statistical heterogeneity with the random-effects variance (2). When few studies were available for an analysis, we used the moment estimates of correlation between sensitivity and specificity in the bivariate model. We calculated positive and negative likelihood ratios (LRs) using the summarized sensitivity and specificity (11, 12). Because studies of a particular biomarker generally used the same definition for a positive test result, we did not plot summary receiver-operating characteristic curves (13). For head-to-head comparisons, we used the same bivariate logistic mixed-effects model, with an added indicator variable for the tests. We conducted analyses for each biomarker by using data from all patients and data stratified according to whether testing was performed for initial diagnosis (evaluation of symptoms) or diagnosis of recurrence (surveillance). We also performed analyses stratified by study design features (such as retrospective or prospective or use of a prespecified threshold to define a positive test result), risk of bias (overall and whether the study performed blinding to the results of the index test), the country in which the study was conducted, and tumor grade and stage (14). We assessed the strength of evidence (SOE) for each body of evidence as high, moderate, low, or insufficient based on aggregate study quality, precision, consistency, and directness. Role of the Funding Source This project was funded under contract HHSA290201200014I from the AHRQ, U.S. Department of Health and Human Services. AHRQ staff assisted in developing the scope and key questions. The AHRQ had no role in study selection, quality assessment, or synthesis. Results The literature flow diagram (Figure 1) summarizes the search and selection of articles. Database searches resulted in 4358 potentially relevant articles. After dual review of abstracts and titles, we selected 262 articles for full-text dual review and determined that 57 studies (in 60 publications) met our inclusion criteria (Appendix Table 1) (15-74). Nineteen studies evaluated quantitative NMP22, 4 evaluated qualitative NMP22, 23 evaluated qualitative BTA, 4 evaluated quantitative BTA, 10 evaluated FISH, 13 evaluated ImmunoCyt, and 1 evaluated Cxbladder. Sample sizes ranged from 26 to 3916, mean age ranged from 54 to 77 years, the proportion of male patients ranged from 57% to 88%, and the proportion diagnosed with bladder cancer ranged from 3% to 81%. Eight studies focused on diagnostic testing for signs and symptoms suggestive of bladder cancer, 16 focused on surveillance of previously treated bladder cancer, and 19 evaluated mixed populations. Forty-three studies were conducted in the United States or Europe. We rated 2 studies as having low risk of bias (20, 21), 3 as having high risk of bias (25, 62, 68), and the remainder as having medium risk of bias. Frequent methodological shortcomings were failure to report blinded interpretation of the reference standard, failure to report enrollment of a random or consecutive sample of patients, or failure to report predefined criteria for a positive test result. Figure 1. Summary of evidence search and selection. * Cochrane Central Register of Controlled Trials and Cochrane Database of Systematic Reviews. Includes prior reports, reference lists of relevant articles, and systematic reviews. Appendix Table 1. Biomarker Study Characteristics Appendix Table 1 Continued Appendix Table 1 Continued Appendix Table 1 Continued Quantitative NMP22 Sensitivity of quantitative NMP22 was 0.69 (95% CI, 0.62 to 0.75), and specificity was 0.77 (CI, 0.70 to 0.83) (19 studies), for a positive LR of 3.05 (CI, 2.28 to 4.10) and a negative LR of 0.40 (CI, 0.32 to 0.50) (Appendix Figure 1). Exclusion of 2 studies that used a cutoff other than >10 U/mL for a positive test result (18, 37) resulted in similar sensitivity and specificity. Diagnostic accuracy was similar for evaluation of symptoms and for surveillance. Excluding 1 study with high risk of bias (68) and restricting the analysis to prospective studies, those conducted in the United States or Europe, or those that used a prespecified threshold for a positive test result had little effect on pooled estimates. Restricting the analysis to 3 studies with blinded reference standard interpretation resulted in higher specificity (0.89 [CI, 0.78 to 0.95]) (15, 42, 58). Appendix Figure 1. Sensitivity and specificity of quantitative NMP22. NMP22 = nuclear matrix protein 22; TN = true-negative; TP = true-positive. Qualitative NMP22 Sensitivity of qualitative NMP22 was 0.58 (CI, 0.39 to 0.75), and specificity was 0.88 (CI, 0.78 to 0.94) (4 studies), for a positive LR of 4.89 (CI, 3.23 to 7.40) and a negative LR of 0.48 (CI, 0.33 to 0.71) (Appendix Figure 2) (20, 21, 23, 37). Restricting the analysis to 2 studies with low risk of bias resulted in similar estimates (sensitivity, 0.53 [CI, 0.29 to 0.75]; specificity, 0.87 [CI, 0.74 to 0.94]) (20, 21). Subgroup and sensitivity analyses were limited by small numbers of studies. Appendix Figure 2. Sensitivity and specificity of qualitative NMP22. NMP22 = nuclear matrix protein 22; TN = true-negative; TP = true-positive. Qualitative BTA Sensitivity of qualit

Treatment of muscle-invasive bladder cancer

There is uncertainty regarding the use of bladder‐sparing alternatives to standard radical cystectomy, optimal lymph node dissection techniques, and optimal chemotherapeutic regimens. This study was conducted to systematically review the benefits and harms of bladder‐sparing therapies, lymph node dissection, and systemic chemotherapy for patients with clinically localized muscle‐invasive bladder cancer. Systematic literature searches of MEDLINE (from 1990 through October 2014), the Cochrane databases, reference lists, and the ClinicalTrials.gov Web site were performed. A total of 41 articles were selected for review. Bladder‐sparing therapies were found to be associated with worse survival compared with radical cystectomy, although the studies had serious methodological shortcomings, findings were inconsistent, and only a few studies evaluated currently recommended techniques. More extensive lymph node dissection might be more effective than less extensive dissection at improving survival and decreasing local disease recurrence, but there were methodological shortcomings and some inconsistency. Six randomized trials found cisplatin‐based combination neoadjuvant chemotherapy to be associated with a decreased mortality risk versus cystectomy alone. Four randomized trials found adjuvant chemotherapy to be associated with decreased mortality versus cystectomy alone, but none of these trials reported a statistically significant effect. There was insufficient evidence to determine optimal chemotherapeutic regimens. Cancer 2016;122:842–51.

Treatment of muscle-invasive bladder cancer: A systematic review: Muscle-Invasive Bladder Cancer Treatment

There is uncertainty regarding the use of bladder‐sparing alternatives to standard radical cystectomy, optimal lymph node dissection techniques, and optimal chemotherapeutic regimens. This study was conducted to systematically review the benefits and harms of bladder‐sparing therapies, lymph node dissection, and systemic chemotherapy for patients with clinically localized muscle‐invasive bladder cancer. Systematic literature searches of MEDLINE (from 1990 through October 2014), the Cochrane databases, reference lists, and the ClinicalTrials.gov Web site were performed. A total of 41 articles were selected for review. Bladder‐sparing therapies were found to be associated with worse survival compared with radical cystectomy, although the studies had serious methodological shortcomings, findings were inconsistent, and only a few studies evaluated currently recommended techniques. More extensive lymph node dissection might be more effective than less extensive dissection at improving survival and decreasing local disease recurrence, but there were methodological shortcomings and some inconsistency. Six randomized trials found cisplatin‐based combination neoadjuvant chemotherapy to be associated with a decreased mortality risk versus cystectomy alone. Four randomized trials found adjuvant chemotherapy to be associated with decreased mortality versus cystectomy alone, but none of these trials reported a statistically significant effect. There was insufficient evidence to determine optimal chemotherapeutic regimens. Cancer 2016;122:842–51.

Urinary Biomarkers for Diagnosis of Bladder CancerA Systematic Review and Meta-analysisUrinary Biomarkers for Diagnosis of Bladder Cancer

Bladder cancer is the fourth most commonly diagnosed cancer in U.S. men and the 10th most commonly diagnosed cancer in U.S. women (1). Standard methods for diagnosis of bladder cancer involve cytologic evaluation of urine, imaging tests, and cystoscopy (2, 3). Because cystoscopy is uncomfortable and costly, alternative diagnostic methods have been sought. Urine-based biomarkers have been developed as potential alternatives or adjuncts to standard tests for the initial diagnosis of bladder cancer or identification of recurrent disease (4). Six urinary biomarkers have been approved by the U.S. Food and Drug Administration (FDA) for diagnosis or surveillance of bladder cancer: quantitative nuclear matrix protein 22 (NMP22) (Alere NMP22 [Alere]), qualitative NMP22 (BladderChek [Alere]), qualitative bladder tumor antigen (BTA) (BTA stat [Polymedco]), quantitative BTA (BTA TRAK [Polymedco]), fluorescence in situ hybridization (FISH) (UroVysion [Abbott Molecular]), and fluorescent immunohistochemistry (ImmunoCyt [Scimedx]). The qualitative NMP22 and BTA tests can be performed as point-of-care tests, and the others are performed in a laboratory. One additional test, Cxbladder (Pacific Edge Diagnostics USA), is a laboratory-developed test that does not require FDA approval. Other biomarkers have been developed but are not FDA-approved. The purpose of this study was to systematically review the evidence on the comparative accuracy of urinary biomarkers for diagnosis of bladder cancer. It was done as part of a larger review (5) on the evaluation and treatment of nonmuscle-invasive bladder cancer that was nominated to the Agency for Healthcare Research and Quality (AHRQ) by the American Urological Association for use in updating its guidelines. Methods Detailed methods and data for this review, including the analytic framework, key questions, search strategies, inclusion criteria, study data extraction, and quality ratings, are available in the full report (5). The protocol was developed using a standardized process (6) with input from experts and the public and is registered in the PROSPERO database (7). This article focuses on the accuracy of urinary biomarkers for initial diagnosis of bladder cancer or for diagnosis of recurrent disease, including any variance in diagnostic accuracy based on tumor characteristics, patient characteristics, or the nature of presenting signs or symptoms. Data Sources and Searches A research librarian searched multiple electronic databases, including Ovid MEDLINE (January 1990 through June 2015), the Cochrane Central Register of Controlled Trials, and the Cochrane Database of Systematic Reviews (through June 2015). We also reviewed reference lists and searched ClinicalTrials.gov. Study Selection Two investigators independently reviewed abstracts and full-text articles against prespecified eligibility criteria. We included cross-sectional and cohort studies on the diagnostic accuracy of urinary biomarkers in adults who had signs or symptoms of bladder cancer or were undergoing surveillance for recurrent disease after treatment. We focused on urinary biomarkers approved by the FDA for the diagnosis of bladder cancer (quantitative or qualitative NMP22, qualitative or quantitative BTA, FISH, and ImmunoCyt) or classified by the FDA as a laboratory-developed test (Cxbladder). We excluded studies that used a casecontrol design; studies that did not evaluate the diagnostic accuracy of biomarkers against standard diagnostic methods (cystoscopy and histopathology); and studies on the accuracy of biomarkers for screening in assessing prognosis, guiding therapy, or monitoring response to treatment. Data Extraction and Quality Assessment One investigator extracted details about the setting, tests evaluated, definition of a positive test result, study design, reference standard, inclusion criteria, population characteristics, proportion found to have bladder cancer, bladder cancer stage and grade, results, and funding sources. We constructed 22 tables with the number of true-positive, false-positive, true-negative, and false-negative results from published sample sizes, prevalence, sensitivity, and specificity. A second investigator verified extractions for accuracy. Two investigators independently assessed the risk of bias for each study as low, moderate, or high using criteria adapted from QUADAS-2 (Quality Assessment of Diagnostic Accuracy Studies 2) (8). Discrepancies were resolved through discussion and consensus. Data Synthesis and Analysis We performed meta-analyses for sensitivity and specificity using a bivariate logistic mixed-effects model (9) with SAS, version 10.0 (SAS Institute) (10). We assumed random effects with a bivariate normal distribution and measured statistical heterogeneity with the random-effects variance (2). When few studies were available for an analysis, we used the moment estimates of correlation between sensitivity and specificity in the bivariate model. We calculated positive and negative likelihood ratios (LRs) using the summarized sensitivity and specificity (11, 12). Because studies of a particular biomarker generally used the same definition for a positive test result, we did not plot summary receiver-operating characteristic curves (13). For head-to-head comparisons, we used the same bivariate logistic mixed-effects model, with an added indicator variable for the tests. We conducted analyses for each biomarker by using data from all patients and data stratified according to whether testing was performed for initial diagnosis (evaluation of symptoms) or diagnosis of recurrence (surveillance). We also performed analyses stratified by study design features (such as retrospective or prospective or use of a prespecified threshold to define a positive test result), risk of bias (overall and whether the study performed blinding to the results of the index test), the country in which the study was conducted, and tumor grade and stage (14). We assessed the strength of evidence (SOE) for each body of evidence as high, moderate, low, or insufficient based on aggregate study quality, precision, consistency, and directness. Role of the Funding Source This project was funded under contract HHSA290201200014I from the AHRQ, U.S. Department of Health and Human Services. AHRQ staff assisted in developing the scope and key questions. The AHRQ had no role in study selection, quality assessment, or synthesis. Results The literature flow diagram (Figure 1) summarizes the search and selection of articles. Database searches resulted in 4358 potentially relevant articles. After dual review of abstracts and titles, we selected 262 articles for full-text dual review and determined that 57 studies (in 60 publications) met our inclusion criteria (Appendix Table 1) (15-74). Nineteen studies evaluated quantitative NMP22, 4 evaluated qualitative NMP22, 23 evaluated qualitative BTA, 4 evaluated quantitative BTA, 10 evaluated FISH, 13 evaluated ImmunoCyt, and 1 evaluated Cxbladder. Sample sizes ranged from 26 to 3916, mean age ranged from 54 to 77 years, the proportion of male patients ranged from 57% to 88%, and the proportion diagnosed with bladder cancer ranged from 3% to 81%. Eight studies focused on diagnostic testing for signs and symptoms suggestive of bladder cancer, 16 focused on surveillance of previously treated bladder cancer, and 19 evaluated mixed populations. Forty-three studies were conducted in the United States or Europe. We rated 2 studies as having low risk of bias (20, 21), 3 as having high risk of bias (25, 62, 68), and the remainder as having medium risk of bias. Frequent methodological shortcomings were failure to report blinded interpretation of the reference standard, failure to report enrollment of a random or consecutive sample of patients, or failure to report predefined criteria for a positive test result. Figure 1. Summary of evidence search and selection. * Cochrane Central Register of Controlled Trials and Cochrane Database of Systematic Reviews. Includes prior reports, reference lists of relevant articles, and systematic reviews. Appendix Table 1. Biomarker Study Characteristics Appendix Table 1 Continued Appendix Table 1 Continued Appendix Table 1 Continued Quantitative NMP22 Sensitivity of quantitative NMP22 was 0.69 (95% CI, 0.62 to 0.75), and specificity was 0.77 (CI, 0.70 to 0.83) (19 studies), for a positive LR of 3.05 (CI, 2.28 to 4.10) and a negative LR of 0.40 (CI, 0.32 to 0.50) (Appendix Figure 1). Exclusion of 2 studies that used a cutoff other than >10 U/mL for a positive test result (18, 37) resulted in similar sensitivity and specificity. Diagnostic accuracy was similar for evaluation of symptoms and for surveillance. Excluding 1 study with high risk of bias (68) and restricting the analysis to prospective studies, those conducted in the United States or Europe, or those that used a prespecified threshold for a positive test result had little effect on pooled estimates. Restricting the analysis to 3 studies with blinded reference standard interpretation resulted in higher specificity (0.89 [CI, 0.78 to 0.95]) (15, 42, 58). Appendix Figure 1. Sensitivity and specificity of quantitative NMP22. NMP22 = nuclear matrix protein 22; TN = true-negative; TP = true-positive. Qualitative NMP22 Sensitivity of qualitative NMP22 was 0.58 (CI, 0.39 to 0.75), and specificity was 0.88 (CI, 0.78 to 0.94) (4 studies), for a positive LR of 4.89 (CI, 3.23 to 7.40) and a negative LR of 0.48 (CI, 0.33 to 0.71) (Appendix Figure 2) (20, 21, 23, 37). Restricting the analysis to 2 studies with low risk of bias resulted in similar estimates (sensitivity, 0.53 [CI, 0.29 to 0.75]; specificity, 0.87 [CI, 0.74 to 0.94]) (20, 21). Subgroup and sensitivity analyses were limited by small numbers of studies. Appendix Figure 2. Sensitivity and specificity of qualitative NMP22. NMP22 = nuclear matrix protein 22; TN = true-negative; TP = true-positive. Qualitative BTA Sensitivity of qualit