Confounding and bias in observational studies in inflammatory bowel disease: a meta‐epidemiological study

• Published in: Alimentary Pharmacology & Therapeutics (Suppl) 2021-03, Vol.53 (6), p.712-721
• Main Authors: Piovani, Daniele, Pansieri, Claudia, Peyrin‐Biroulet, Laurent, Danese, Silvio, Bonovas, Stefanos
• Format: Article
• Language: English
• Subjects: Bias; Case-Control Studies; Colitis; Epidemiologic Studies; Epidemiology; Gastroenterology; Humans; Inflammatory bowel disease; Inflammatory Bowel Diseases; Inflammatory Bowel Diseases - diagnosis; Inflammatory Bowel Diseases - epidemiology; Intestine; Life Sciences; Observational studies; Observational Studies as Topic
• ISSN: 0269-2813; 0953-0673; 1365-2036
• DOI: 10.1111/apt.16222

Summary Background Observational research concerning inflammatory bowel disease (IBD) is highly susceptible to spurious findings because of confounding and bias. Aim To investigate how these issues were reported in this research field. Methods We identified and appraised a random sample of 160 observational studies concerning IBD published in high‐impact gastroenterology journals and the most respected specialty journals of the condition. We applied a standardised methodology to assess how confounding and bias were reported and discussed, and investigated the association between yearly citations and study characteristics using mixed‐effect multivariable regression analysis. Results The authors of 67 out of 160 articles (41.9%) mentioned confounding, and in 89 cases (55.6%) reported any bias. Although most authors applied strategies to minimise confounding or bias (n = 139; 86.9%) and acknowledged at least one unadjusted confounder (n = 116; 72.5%), a minority commented about whether the main findings could have been affected (n = 60; 37.5%). Very few authors (n = 7; 4.4%) called for caution in interpreting the results in the discussion. Reporting of confounding and bias was particularly lacking for case‐control studies, those not using routinely collected data, those employing laboratory analyses as the primary method of assessment and studies investigating non‐modifiable exposures. In adjusted analyses, mentioning or alluding to confounding was positively associated with yearly citations (P = 0.010), whereas calling for a cautious interpretation of the findings was not. Conclusions Reporting of confounding is inadequate and its acknowledgement is often neglected in interpreting high‐impact observational research in IBD. These results encourage a more careful evaluation of the consequences of confounding and bias.