Poster Session II - A223 STUDIES BASED ON HEALTH ADMINISTRATIVE DATA REGARDING RARE OUTCOMES IN INFLAMMATORY BOWEL DISEASE SIGNIFICANTLY UNDERESTIMATE THE TRUE RISK - THE IMPORTANCE OF SPECIFICITY

TL;DR

This study shows that health administrative data significantly underestimates the risk of rare outcomes in inflammatory bowel disease due to misclassification bias.

Contribution

The study introduces the use of quantitative bias analysis to correct for misclassification in health administrative data for IBD research.

Findings

Bias-adjusted relative risks were two to three times higher than reported results in some studies.

The Canadian study showed the largest divergence after bias adjustment.

Most HAD-based IBD studies significantly underestimate true risk due to information bias.

Abstract

Health administrative data (HAD) has increased our knowledge of rare outcomes in inflammatory bowel disease (IBD), such as cancer and mortality. However, most neglect to address the vulnerability to systematic misclassification of IBD diagnosis (information bias)

We aimed to assess the information bias imposed by misclassification of the IBD diagnosis in HAD studies by performing quantitative bias analysis (QBA).

We identified pediatric-onset IBD (PIBD) HAD studies assessing cancer risk in which the PIBD case identification was based on published validation studies. We then performed probabilistic QBA to adjust for non-differential exposure misclassification using the sensitivity and specificity values from country or region-specific validation studies.

We included four studies presented in Table 1, along with reported and bias-adjusted outcome estimates. Of these studies, the Canadian and Danish studies, which had specificity values of 0.92 and 0.91, respectively, had the largest degree of misclassification bias, biasing the results towards the null. The bias-adjusted relative risks were two to three times higher than the reported results. The most considerable divergence was found in the Canadian study, with a reported relative risk of 2.0 (95%CI: 1.2-3.4) and a bias-adjusted relative risk of 5.8 (95%CI: 2.5-13.7). The corresponding RD was 1.0% (95%CI: 0.1-1.9) and the bias-adjusted RD was 3.8% (95%CI: 1.4-7.9).

HR, Hazard ratio. RR, relative risk. RD, risk difference.

We exemplify the importance of supplementing HAD-based cohort studies with QBA. To do this, it is imperative that validation studies are available on case assertion algorithms within the specific registry for any disease group that a forthcoming study would be reporting on. Our results indicate that most HAD-based studies on rare long-term consequences of IBD significantly underestimate the true risk of the outcomes. Because the same underlying information bias exists in adult literature, our result can be extrapolated to HAD-based IBD studies in all age groups.

None