Michael Solomon, MB, BCh, MSc, FRACS; Toufic El-Khoury, MB, BS, FRACS
What is the diagnostic accuracy of alarm features for colorectal cancer?
Included studies compared individual alarm features, statistical models, or both, with lower gastrointestinal (GI) investigational tests (colonoscopy, barium enema, or computed tomographic colography, or flexible sigmoidoscopy alone with ≥ 1 y follow-up or data on potentially missed colorectal cancers at study end) for diagnosing colorectal cancer in unselected patients > 16 years of age who had lower GI symptoms. Symptoms had to be recorded before tests were done, and > 90% of patients had to undergo the tests. Outcomes were sensitivity, specificity, and positive and negative likelihood ratios (LRs).
MEDLINE (1950 to Oct 2007), EMBASE/Excerpta Medica (1988 to Oct 2007), CINAHL (1982 to Oct 2007), and reference lists were searched for cross-sectional studies that included > 100 patients and reported > 1 colorectal cancer diagnosis. Authors were contacted for additional data. 15 studies (n = 19 443) met the selection criteria: 13 evaluated individual alarm features (rectal bleeding, change in bowel habit, anemia, weight loss, diarrhea, or abdominal mass), 1 evaluated a statistical model, and 1 did both. 2 studies were done in primary care settings, 11 in secondary care settings, 1 in both settings, and 1 was population-based. 4 studies did independent blinded comparisons of alarm features with valid reference standards in consecutive patients (quality level 1 or 2 of 5); 11 studies were low quality (level 4 of 5).
The pooled prevalence of colorectal cancer was 6%. Pooled specificity was ≥ 80% for 6 of 8 individual alarm features; sensitivity was low for all individual features (Table). Pooled sensitivity for statistical models was 90% and specificity was 59% (Table).
Individual alarm features have moderate-to-high specificity and low sensitivity for diagnosing colorectal cancer; palpable abdominal masses and dark-red rectal bleeding have high specificity. Statistical models have low specificity and high sensitivity.
Pooled test characteristics of individual alarm features or statistical models for diagnosing colorectal cancer*
*Diagnostic terms and abbreviations defined in Glossary.
Solomon M, El-Khoury T. Review: Individual alarm features have moderate-to-high specificity and low sensitivity for diagnosing colorectal cancer. Ann Intern Med. 2009;150:JC3–14. doi: 10.7326/0003-4819-150-8-200904210-02014
Download citation file:
Published: Ann Intern Med. 2009;150(8):JC3-14.
Colorectal Cancer, Gastroenterology/Hepatology, Gastrointestinal Cancer, Hematology/Oncology.
Copyright © 2018 American College of Physicians. All Rights Reserved.
Print ISSN: 0003-4819 | Online ISSN: 1539-3704
Conditions of Use