Accuracy of Fecal Immunochemical Tests for Colorectal Cancer: Systematic Review and Meta-analysis

By Jeffrey-K Lee, Elizabeth G. Liles, S. Bent, T. Levin, D. Corley. Annals of Internal Medicine, 2014

675 citations recorded by the source corpus.

Colorectal cancer (CRC) is the second-leading cause of cancer-related deaths in the United States (1). Randomized, controlled trials have shown that annual or biennial fecal occult blood tests (FOBTs) are associated with a 15% to 33% decrease in CRC mortality rates (24). However, FOBTs only detect approximately 13% to 50% of cancer with 1 round of screening in asymptomatic patients (5, 6). In addition, adherence to repeated rounds of FOBTs in real-world screening programs is low, raising concern about their effectiveness as screening tests (7, 8). Fecal immunochemical tests (FITs) are more sensitive at detecting both CRC and adenomas than FOBTs (9, 10). Many FITs require only 1 or 2 stool samples, and none require dietary or medication restrictions, increasing ease of use. In 2008, several U.S. professional societies endorsed the use of FITs to replace FOBTs because of the former's improved performance characteristics and potential for higher participation rates (10, 11). Countries in Europe and Asia have also adopted widespread CRC screening programs using FITs (12, 13). However, the diagnostic characteristics of these tests have been difficult to estimate, with reported sensitivities ranging from 25% to 100% for CRC and specificities usually exceeding 90% (9, 14, 15). The lack of a precise estimate of sensitivity has resulted in confusion among health care providers about the sources of this variation, how best to apply FITs for CRC screening, the optimal number of stool samples for testing, optimal cutoff value for a positive test result, and whether any brand of FIT is superior to others. Our analysis provides a quantitative meta-analysis of the diagnostic accuracy (sensitivity and specificity) of FITs for CRC. In addition, we explored potential sources of heterogeneity by analyzing subgroups classified by FIT sample number, cutoff value for a positive test result, FIT brand, and the reference standard. Methods We developed a protocol on the basis of standard guidelines for the systematic review of diagnostic studies (16, 17) and the strategy used for the U.S. Preventive Services Task Force review in 2008 (9). We followed the STARD (Standards for the Reporting of Diagnostic Accuracy Studies) (18) and Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) (19) statements for reporting our systematic review. This study was conducted as part of the National Cancer Institutefunded consortium, Population-Based Research Optimizing Screening through Personalized Regimens. The overall aim of this consortium is to conduct multisite, coordinated, transdisciplinary research to evaluate and improve cancer screening processes. Data Sources and Searches We included all studies identified in the previous USPSTF report (9) plus other studies identified by a search of FIT for CRC between 1 January 2008 and 31 August 2013 using MEDLINE (via Ovid), EMBASE, Database of Abstracts of Reviews of Effects, Health Technology Assessment Database, Cochrane Database of Systematic Reviews, and Cochrane Central Register of Controlled Trials. We also searched bibliographies and reference lists of eligible papers and related reviews, consulted experts in the field, and contacted several authors from the included studies to locate additional studies. The Appendix Table 1 provides further details of our search strategy. Appendix Table 1. Search Strategy Study Selection Two persons independently reviewed the pertinent studies to determine eligibility. We included studies if they met all of the following criteria: evaluated the diagnostic accuracy of FITs for CRC; reported absolute numbers of true-positive, false-negative, true-negative, and false-positive observations, or if these same variables could be obtained from personal communication; used a randomized trial or cohort study design; evaluated adult participants who were asymptomatic and older than 18 years with a mean age greater than 40 years; and reported an appropriate reference standard (colonoscopy or 2-year longitudinal follow-up of controls with medical records or cancer registry). Given that only a subset of studies reported data on adenomatous polyps and that there is variability in definitions of polyps, we limited the scope of this analysis to test performance characteristics for detecting CRC; we excluded studies reporting test performance estimates for detection of adenomas only. We did not include conference abstracts and casecontrol studies, which, by creating spectrum bias, can overestimate the accuracy of a test (20). To avoid duplicate reporting of the same population for studies reporting several cutoff values or numbers of samples, we used the cutoff value or sample number most commonly used in current practice in the United States, used in national recommendations, or recommended by expert opinion in the main analyses. In addition, we selected the sample number or cutoff value a priori that was most similar to those in other studies for our subgroup analyses. Data Extraction and Quality Assessment Two reviewers independently evaluated and extracted relevant information from each included study and assessed study quality via the Quality Assessment of Diagnostic Accuracy Studies 2 instrument (21). For studies with incomplete or unavailable information, we contacted the corresponding authors or coauthors to complete missing information. Of the 15 contacted authors, 12 provided additional data. We converted units for cutoff thresholds for a positive test result in each study to micrograms of hemoglobin per gram of stool, as recommended by leading experts (22). Data Synthesis and Analysis We calculated the sensitivity, specificity, positive predictive value, negative predictive value, positive likelihood ratio (LR), and negative LR with 95% CI of each study. A positive LR greater than 5 and a negative LR less than 0.2 provide strong diagnostic evidence to rule in or rule out diagnoses, respectively (23). The overall pooled sensitivity and specificity of FIT for CRC were estimated using a bivariate random-effects model (24). We calculated the pooled positive LR and negative LR along with the respective CI using the bivariate model (24) according to the method used by Zwinderman and colleagues (25). We also generated a hierarchical summary receiver-operating characteristic curve that plots the individual and summary estimates of sensitivities and specificities along with a 95% confidence and prediction region (26). Last, we calculated the area under the hierarchical summary receiver-operating characteristic curve. An area under the curve between 0.9 and 1.0 indicates that the test in question is highly accurate (27). The Q value and the inconsistency index (I 2) test were used to estimate the heterogeneity between each study (28). We regarded values of 25%, 50%, and 75% for the I 2 as indicative of low, moderate, and high statistical heterogeneity, respectively (28). In addition, we calculated the between-study variance of logit sensitivity and logit specificity (24, 29). In diagnostic accuracy studies, 1 of the primary causes of heterogeneity is the threshold effect, which occurs when different cutoff values are used between studies to define a positive (or negative) test result. We searched for evidence of a threshold effect by calculating the squared correlation coefficient estimated from the between-study covariance variable in the bivariate model (30). We stratified studies into 4 subgroups on the basis of the number of FIT samples (1, 2, or 3 samples), prespecified cutoff values of fecal hemoglobin concentration for a positive test result (<20 g/g, 20 to 50 g/g, and >50 g/g), brand, and reference standard used to follow up on patients with negative FIT results. Cutoff values were grouped to ensure an adequate number of data sets for each analysis. To determine whether studies using older (discontinued) FITs were causing heterogeneity in our summary estimates, we did sensitivity analyses by removing these studies and recalculating the I 2 test for the remaining group. In addition to threshold effect and subgroup analyses, we did a bivariate random-effects meta-regression analysis to identify additional sources of heterogeneity that may have influenced our overall summary estimates (30). We used the following prespecified variables for our meta-regression: type of FIT (qualitative, point-of-care tests or quantitative, automated tests), geographic region (Asian or non-Asian countries), and enrollment of patients younger than 40 years. We used Stata, version 12.0 (StataCorp, College Station, Texas), for all statistical analyses. All tests were 2-sided, and we considered P values less than 0.05 to be statistically significant. Role of the Funding Source The study was funded by the National Institute of Diabetes and Digestive and Kidney Diseases and the National Cancer Institute. The funding source had no role in the conception, design, analysis, or conduct of the review. Results Study Selection The 2008 USPSTF report (9) included 9 studies in its systematic review (3139); our literature search identified 1771 additional new potential sources (Figure 1). After abstract review, we identified 53 articles for full-text review; of these, 18 unique articles satisfied all inclusion criteria and were included in our analysis (14, 15, 3146). Because 1 article (46) evaluated more than 1 FIT brand in a head-to-head comparison, the final analysis included 19 studies or data sets. Figure 1. Summary of evidence search and selection. USPSTF = U.S. Preventive Services Task Force. Characteristics of Included Studies Table 1 and the Supplement show the main characteristics of the included studies. Eighteen articles described 19 cohort studies of FIT sensitivity and specificity for CRC in average-risk asymptomatic patients; sample sizes ranged from 80 to 27860. Twelve studies (14, 3336, 4042, 4446) used colonoscopy in all patients, regardless of FIT results, as the re

Subfields: Epidemiology, Oncology, Surgery

View Paper: https://www.semanticscholar.org/paper/97ee62b7a34177bdce898f08dfed07e4ca3e05ee

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Oncology

  1. Douglas Rex

Citations in the corpus