Letter on ‘Meta‐Analysis: Mortality Trends and Risk Factors in Severe Alcohol‐Associated Hepatitis’

We read with great interest the meta-analysis by Siddique et al. [1], which provides a timely synthesis of short-term mortality in severe alcohol-associated hepatitis (sAH). The application of both frequentist and Bayesian frameworks is a notable strength. However, we have profound concerns regarding the analytical approach to the primary conclusion that mortality has not improved in recent decades, which may be substantially flawed. The authors' conclusion of no credible mortality improvement rests heavily on the non-significance of publication year as a covariate in their meta-regression. This model is critically compromised by the systematic and profound structural heterogeneity of the ‘standard care’ groups across decades. As illustrated in your table 1, the control arm for a 1978 trial was ‘diet’, while in 2023, it constituted ‘prednisolone’ combined with advanced prophylactic antibiotics [2, 3]. Pooling these fundamentally different clinical realities into a single ‘standard care’ cohort and then testing for a time trend creates a misleading null signal. The observed decline in 28-day mortality from > 50% to ~25%, as shown in your Bayesian analysis, likely represents the genuine, cumulative benefit of evolving standard care. To assert no improvement is to ignore this evident clinical progress captured by your own data. Furthermore, the model selection strategy, which prioritised the corrected Akaike Information Criterion (AIC), is inappropriate for this context. The AIC favours parsimony but cannot distinguish a biologically implausible model from a correct one. Your final model identifies ‘follow-up duration’ as the principal mortality determinant—a tautological finding, as the probability of death intrinsically increases over time in a fatal disease. That this model was selected over one containing ‘treatment era’ underscores the limitation of a purely algorithmic approach over clinical reasoning. Finally, the assertion of a significant association between MELD score and mortality (β = +0.20 per point, p = 0.037) is not robust. This analysis was based on a small, non-random subset of only 12 studies reporting MELD scores, heavily predisposed to publication and reporting biases. The conclusion lacks the reliability necessary to inform clinical prognostication, a point not adequately emphasised. We believe a re-analysis comparing mortality between pre-specified, clinically distinct eras (e.g., pre- and post-corticosteroid standardisation) or using a network meta-analysis structure to account for evolving standard care, would provide a more valid assessment of temporal trends. In this context, recent reviews and consortium studies—such as those summarising current and emerging therapies for acute alcohol-associated hepatitis, outcomes of patients with sAH and acute kidney injury, interventions for alcohol use disorder in cirrhotic or sAH patients, and current criteria for early liver transplantation in sAH—collectively highlight the evolving management landscape and reinforce the need for era-stratified analyses [4-7]. The current conclusions risk misdirecting clinical and research priorities by underestimating the value of established supportive care. Chao Zhao: writing – original draft, writing – review and editing. The author declares no conflicts of interest. This article is linked to Siddique et al. paper. To view this article, visit https://onlinelibrary.wiley.com/doi/10.1111/apt.70383. The data that support the findings of this study are available from the corresponding author upon reasonable request.