Performance Characteristics of Fecal Immunochemical Tests for Colorectal Cancer and Advanced Adenomatous Polyps

Colorectal cancer (CRC) is a leading cause of death among digestive diseases and the second leading cause of cancer-related death in the United States (1). Despite the effectiveness and cost-effectiveness of screening (24), only 60% to 65% of the eligible population is current with screening (5), a rate that has fallen short of the goal of 80% by 2018 (2, 5, 6). This reflects concerns over the best test and strategy for screening. Colonoscopy is the most frequently used screening test in the United States (5), but several other countries use annual or biennial stool blood tests or a combination of stool testing and lower endoscopy (7, 8). Although studies have shown that guaiac-based fecal occult blood testing reduces CRC incidence and mortality (913), it has several shortcomings, including low single-application sensitivity for CRC, poor detection of advanced adenomas (those with a diameter 1 cm, villous histologic characteristics, or high-grade dysplasia), the need for dietary and medication restrictions, and the requirement for more than 1 specimen. Use of the fecal immunochemical test (FIT) for human globulin is more sensitive and specific than guaiac-based fecal occult blood testing for CRC and advanced adenomas and has higher rates of participation and acceptance (1416). However, studies evaluating FIT performance characteristics have shown inconsistent findings for CRC and advanced adenomas. A systematic review published in 2014 summarized performance characteristics for CRC (17) but not for advanced adenomas. The objectives of this systematic review and meta-analysis were to provide an updated summary of FIT performance for CRC, quantify FIT performance characteristics for advanced adenomas, and evaluate whether variation in reported performance characteristics among studies is a function of the threshold used to define a positive test result or the test brand. Methods Rather than developing and registering a new formal protocol, we used 2 prior systematic reviews as guides for our study methods (17, 18). We followed standard procedures for systematic reviews and reported results according to the PRISMA (Preferred Reporting Items for Systematic reviews and Meta-Analyses) guidelines (19, 20). Data Sources and Searches We performed English-language searches of the Ovid MEDLINE, PubMed, Embase, and Cochrane databases from inception to 17 October 2018 to identify studies assessing performance of 1 or more FITs. Searches were done using various combinations of the following terms: feces, occult blood, colon cancer, cancer screening, early diagnosis, immunochemistry, and FIT (Appendix Table 1). We also reviewed reference lists of relevant systematic reviews and meta-analyses (17, 18) and of articles that met selection criteria. Appendix Table 1. Detailed Search Strategies Study Selection Two authors independently screened all titles and potentially relevant abstracts and then full texts of potentially eligible articles. Prospective or retrospective observational studies that were published in English; evaluated FIT sensitivity and specificity in asymptomatic, average-risk adults; and used colonoscopy as the reference standard were eligible for inclusion. Data only in abstract form or gray literature were not eligible. Data Extraction and Quality Assessment Two reviewers reviewed descriptive and quantitative data from each study. Data extraction was done primarily by 1 author (R.N.G.) and was independently validated by 2 others (T.F.I. or T.W.E.). For each study, we extracted data on sample size, mean age, brands of FIT used, thresholds for positivity (expressed as micrograms of hemoglobin per gram of feces), numbers of participants with CRC and advanced adenomas, and performance characteristics for both. Raw data on CRC and advanced adenomas were extracted when available. When only computed data were available, raw data were calculated based on identified proportions. For articles with missing data, the corresponding authors were contacted. When more than 1 FIT cutoff or threshold was used, performance characteristics for commonly used thresholds were extracted. Two authors (among T.F.I., R.N.G., and T.W.E.) independently assessed study characteristics and evaluated study quality by using the revised version of the QUADAS-2 (Quality Assessment of Diagnostic Accuracy Studies 2) tool (21). Discrepancies between reviewers for study quality assessment were resolved by discussion. The authors did not grade the quality of their own studies. Data Synthesis and Analysis Sensitivity and specificity at 1 or more thresholds were reported for each study. We combined studies and report results for CRC and advanced adenomas, with studies grouped by threshold (10 to 20 g/g). To assess statistical heterogeneity, we calculated the I 2 measure, which indicates the percentage of total variation across studies that is due to heterogeneity rather than chance (22). For all summary-level estimates, we used a bivariate generalized linear mixed model to simultaneously estimate pooled measures of sensitivity and specificity separately for CRC and advanced adenomas while accounting for the potential correlation between sensitivity and specificity. The bivariate approach provides unbiased estimates of sensitivity, specificity, and their correlation (23) and does not rely on an ad hoc continuity correction for zero marginal counts. Likelihood ratios were calculated using the bivariate estimates; positive likelihood ratios were calculated as sensitivity divided by (1 specificity), and negative likelihood ratios were calculated as (1 sensitivity) divided by specificity. Summary receiver-operating characteristic curves were obtained along with 95% confidence regions for the bivariate estimates of sensitivity and (1 specificity). We also combined studies by FIT brand and by threshold to enable indirect comparisons. Meta-DiSc software (Hospital Universitario Ramn y Cajal) (24) was used to calculate the I 2 measure. For all other summary estimates of performance characteristics, the glmer function (25) of the lme4 package (26) in R (R Foundation for Statistical Computing) (27) was used to estimate the bivariate generalized mixed models (details are provided in the Supplement). Supplement. Technical Appendix Role of the Funding Source The study was funded by the Department of Medicine and the Melvin and Bren Simon Cancer Center at the Indiana University School of Medicine and by the Regenstrief Institute. The funding sources had no role in the conception, design, or conduct of the study; analysis of the data; review of the manuscript; or the decision to submit it for publication. Results Our search strategy (Appendix Table 1) generated 4976 citations, among which 31 articles were included in the analysis (Figure 1; Appendix Table 2) (2858). Studies were published between 2001 and 2018. Funding sources were not reported for 10 studies and included federal government (n= 15), private or intramural (n= 4), and corporate (n= 2) sources for the other studies. The total number of participants was 120255, and sample sizes ranged from 284 to 21805. Mean age ranged from 48.2 to 64 years. All study populations comprised asymptomatic, mostly average-risk persons in the screening age range (generally 50 to 75 years) who enrolled in screening colonoscopy programs for cancer prevention. Persons with prior colorectal neoplasia, inflammatory bowel disease, high-risk family history, or colonoscopy in the previous 5 to 10 years were excluded, as were (post hoc) those with inadequate bowel preparation or incomplete colonoscopy. Figure 1. Evidence search and selection. FIT= fecal immunochemical test. Appendix Table 2. Characteristics of Included Studies Eighteen FITs were tested, with individual studies testing 1 to 6 FITs. OC-Sensor (Eiken Chemical) was tested in 14 (58%) studies (35, 3942, 4649, 51, 52, 54, 56, 57), including OC FIT-CHEK (Eiken Chemical) in 2 (48, 57) of those 14 studies, OC-Light (Eiken Chemical) in 5 studies (29, 36, 37, 43, 45), and OC-Hemodia (Eiken Chemical) and FOB Gold (Sentinel Diagnostics) in 3 studies each (28, 30, 32, 33, 55, 58). Many of the remaining FITs are or were available only in a single country or region. Thresholds for positivity ranged from 2 to 67 g/g, with 11 studies using a positivity threshold of less than 10 g/g, 17 using a threshold of 10 g/g, 9 using a threshold of 11 to 19 g/g, and 26 using a threshold of 20 g/g or greater. Performance characteristics of FITs for CRC and advanced adenomas based on each threshold tested for individual studies are shown in Appendix Table 2. All studies assessed the sensitivity and specificity of 1 or more FITs for advanced adenomas, which ranged in prevalence from 1.26% to 12.2%, and all but 3 studies (33, 39, 51) did so for CRC, which ranged in prevalence from 0.15% to 3.48%. We judged the quality of most studies as high (Appendix Figure 1). All were cross-sectional. Only 1 had a prospective casecontrol design (46); for this study, we used data only from the control group, which, as in the other studies, comprised persons having screening colonoscopy. A lack of detail in study methods for nearly half the studies precluded us from knowing whether a consecutive or random sample of persons participated. Despite this, we assessed most studies as having low to moderate risk of selection bias. For approximately 30% of the studies, it was unclear whether FIT results and colonoscopic findings were interpreted independently. In 25% of the studies, the interval between the FIT and colonoscopy was not specified. Appendix Figure 1. QUADAS-2 study quality stacked bar charts. Q = question; QUADAS-2 = Quality Assessment of Diagnostic Accuracy Studies 2. The studies tested 1 or more FITs; several used more than 1 threshold with the objective of determining the optimal one. The FIT was done before colonoscopy in all but 1 study (49), which provided no information. Although 5