Letter to the editor: Focus on metabolic dysfunction-associated fatty liver disease, even if patients have a history of excessive alcohol consumption.

To the editor, Recently, we followed the study on metabolic dysfunction–associated fatty liver disease (MAFLD) and excessive alcohol consumption published in hepatology by van Kleef et al.1 They pointed out that MAFLD and excessive alcohol consumption were independent risk factors for mortality. The authors made a multidimensional validation and a meticulous subgroup analysis of these findings in a cohort containing 12,656 participants. The contribution of van Kleef et al. in promoting the concept of MAFLD and identifying the unique role of MAFLD in increasing mortality is highly appreciated. However, we believe several issues deserve further discussion. First, the authors used multivariate Cox regression to determine the independent role of MAFLD and excessive alcohol consumption in increasing mortality (their Table 2). It is unclear how the authors selected the variables. The previous study had shown that a directed acyclic graph (DAG) was recommended when screening for variables.2Figure 1A suggests that MAFLD might be a mediating factor of excessive alcohol consumption. Including MAFLD in the analysis of mortality might obscure the total effect of excessive alcohol consumption, which resulted in an underestimation of the value of the effect of excessive alcohol consumption. However, when studying the effect of MAFLD on mortality, excessive alcohol consumption could be included as a confounding factor in a multifactorial Cox regression analysis. We recommend that the authors attempt to add multifactorial Cox regression analyses that do not include MAFLD to determine the total effect value of excessive alcohol consumption. Meanwhile, a DAG showing the screening process for all variables could add to the scientific validity and credibility of the study.FIGURE 1: Diagram of the DAG and multistate model. (A) The DAG of excessive alcohol consumption and MAFLD. (B) Potential five‐states model of excessive alcohol consumption and patients with MAFLD. The transient state includes healthy, excessive alcohol consumption and MAFLD. The absorbing state includes liver‐related death and non‐liver‐related death.The authors mentioned that the study did not distinguish between all‐cause and liver‐related deaths. While this did not affect the reliability of the conclusions, it hindered researchers from gaining insight into the relationship between excessive alcohol consumption, MAFLD, and mortality. Multistate models are ideal when examining the association between multiple states and outcome events.3 We can calculate the probability of transfer between different states and predict the final prognosis of patients by synthesizing these probabilities (Figure 1B). Multistate models construct survival analysis models that are more consistent with real‐world association patterns. The construction of a multistate model relies on complete follow‐up data. Therefore, more detailed tracking and documentation of different states are necessary when constructing study cohorts. We call on researchers to experiment with DAG and multistate models for variable screening and survival analysis to optimize the research process.