Oral Health-Head and Neck Cancers: Addressing Confounding Through Negative Control and Quantitative Bias Analyses.

Abstract

Objectives: While there are plausible biological explanations for the association between oral health and head and neck cancers (HNC), existing studies have yielded conflicting results. A key concern is that these associations are influenced by mediators, unmeasured risk factors, and biases. To address this, a negative control exposure was used to evaluate whether the associations between oral health and HNC risk could be attributed to unmeasured confounding. Additionally, quantitative bias analysis (QBA) was performed to estimate the extent of non-differential misclassification of exposure.

Methods: The HeNCe study, a hospital-based case-control study, recruited incident HNC cases (n = 389) frequency matched to controls (n = 429) by sex and age (within 5 years) from four major referral hospitals in Montreal, Canada. In-person interviews collected information on life course exposures. Unconditional logistic regression estimated the odds ratios (OR) and 95% confidence intervals (CI) for the associations between oral health indicators and HNC, controlling for confounders identified using directed acyclic graphs (DAG). Sexually transmitted diseases (STD) were used as a negative control exposure to test for unmeasured confounding in the associations. QBA, using predetermined bias parameters from previous studies, estimated the magnitude and direction of exposure misclassification bias.

Results: Complete denture use and having more than nine missing teeth were associated with an increased HNC risk [OR = 1.33, 95% CI (0.93-1.90) & OR = 1.31, 95% CI (0.93-1.83)], respectively. Similar results were obtained when stratified by HNC subsite. Negative control analysis yielded a null finding, indicating no significant bias due to unmeasured confounders. Bias-corrected estimates of the association between oral health indicators and HNC risk moved further from the null.

Conclusion: Negative control exposure analysis indicated that unmeasured confounding did not affect the association between oral health and HNC risk. QBA yielded corrected estimates of increased magnitude, suggesting that the crude associations may have been underestimated.