A Principled Framework for Mendelian Randomization in Oral Health Research

Abstract

In recent years, the popularity of Mendelian randomization (MR) as a technique to investigate causal relationships has grown exponentially. This approach leverages randomness inherent in the allocation of genetic variants to circumvent issues of unobserved confounding and reverse causation that compromise many causal claims based on observational data. In this manuscript, the key assumptions required for MR analyses to yield credible causal inferences are reviewed. It is emphasized that even technically rigorous MR analyses may yield implausible conclusions if these assumptions are violated. The goal is to provide readers from nonstatistical backgrounds with a clear understanding of the foundational assumptions underpinning MR, complemented by a detailed discussion of specific considerations in oral health research, particularly when periodontitis is used as the exposure or outcome. As with any study design, MR is appropriate and can provide compelling evidence for causality in certain contexts, while its application may be unwarranted in others. This methodological article aims to equip readers with the tools to critically assess MR findings, distinguish scenarios where its use is justified, and understand the key steps for correct execution and reporting. A checklist for assessing the credibility of MR investigations is also provided.

Keywords: causal inference; dentistry; epidemiology; mendelian randomization; periodontal diseases; randomized controlled trial.

Figure 1. Directed acyclic graph representing the causal relationships in a hypothetical study investigating the effect of fluoride levels on dental caries.

Arrows indicate that one variable causally influences another, with the direction of the arrow representing the causal relationship. The causal effect of interest is labelled in the diagram.

Figure 2

Directed acyclic graph representing the causal relationships in a hypothetical MR study investigating the effect of smoking on periodontitis. The three core assumptions required for valid MR analyses are labelled in the diagram: (1) The effect of the genetic variant on the exposure (relevance). (2) No effect of the confounders on the genetic variant (exchangeability). (3) No direct effect of the genetic variant on the outcome (exclusion restriction).

Figure 3

Schematic diagram showing the analogy between a simple two-armed randomized clinical trial, and an MR analysis with a dichotomous genetic variant as the instrumental variable. In both cases, under the instrumental variable assumptions, an association between the outcome and either treatment allocation (in a randomized clinical trial) or the genetic variant (in MR) is indicative of a causal effect of the exposure on the outcome.

Figure 4

Directed acyclic graph illustrating pleiotropy in a hypothetical MR study investigating the effect of body mass index (BMI) on periodontitis. Vertical pleiotropy does not cause bias in the MR analysis as inflammation is on the causal pathway from BMI to periodontitis. Horizontal pleiotropy does cause bias as type 2 diabetes mellitus (T2DM) has effects on periodontitis via a pathway distinct from BMI, so the instrument is in violation of the exclusion restriction assumption.

Figure 5. Three categories of MR investigations: plausible, questionable, and implausible.