Dear Editor, We read with great interest the recent study by Liu et al published in the International Journal of Surgery, entitled “Coronavirus disease 2019-related risk factors for postoperative delirium: a secondary analysis of an observational cohort study”1. Using a rigorous prospective design and the validated 3D-CAM assessment tool, the authors provide the first systematic evidence that a diagnosis of COVID-19 within seven weeks before surgery – particularly symptomatic infection – is significantly associated with an increased risk of postoperative delirium (POD), with a notably stronger effect observed in patients younger than 65 years. These findings offer valuable clinical guidance for perioperative neurocognitive risk stratification. The letter to the editor is compliant with the TITAN Guidelines 2025 – governing declaration and use of artificial intelligence2. Although the study is methodologically robust and employs a standardized delirium assessment, certain limitations merit consideration. First, the definition of SARS-CoV-2 exposure relied on clinical diagnosis or patient self-report without uniform laboratory confirmation such as PCR or serological testing, which may introduce exposure misclassification, particularly among individuals with asymptomatic or mild infections. Second, the analysis did not fully account for baseline preoperative cognitive function – a well-established major predictor of POD. Given that prior SARS-CoV-2 infection can lead to persistent cognitive symptoms (“brain fog”), any observed association might partly reflect residual confounding by subclinical cognitive impairment rather than a direct pathogenic effect of the virus3. While the authors adjusted for several covariates, such unmeasured or imperfectly measured confounders remain difficult to eliminate entirely within an observational framework. Consequently, the fundamental question of whether SARS-CoV-2 infection causally increases the risk of POD remains unresolved due to potential confounding and reverse causation inherent in observational designs. To address this, we conducted a two-sample Mendelian randomization (MR) analysis using publicly available genome-wide association study (GWAS) summary statistics4, 5. We used genetic instruments for susceptibility to COVID-19 (GWAS ID: ebi-a-GCST010776) as the exposure and delirium not induced by alcohol or other psychoactive substances from the FinnGen consortium (outcome ID: finn-b-F5DELIRIUM) as the outcome. In the primary inverse-variance weighted analysis, genetically predicted liability to COVID-19 was significantly associated with a higher risk of delirium (odds ratio OR = 1. 20, 95% confidence interval CI: 1. 02 to 1. 41; P = 0. 024) (Fig. 1). This finding was corroborated by the Bayesian weighted MR (BWMR) method6, which yielded a consistent estimate (OR = 1. 21, 95% CI: 1. 02 to 1. 44; P = 0. 026) (Fig. 1). Although the MR-Egger regression did not yield a statistically significant causal estimate (P = 0. 445), the direction of effect was consistent with the primary analyses, and the MR-Egger intercept test provided no evidence of directional horizontal pleiotropy (intercept P > 0. 05) (Fig. 1). This conclusion was further reinforced by complementary sensitivity analyses: leave-one-out permutation demonstrated that no single instrumental variant disproportionately influenced the overall effect, and funnel plot symmetry indicated the absence of substantial heterogeneity or publication bias (Fig. 1). Figure 1.: Mendelian randomization estimates of the causal association between genetically predicted susceptibility to SARS-CoV-2 infection and risk of delirium. Our MR results provide genetic evidence that the association reported by Liu et al is unlikely to be explained solely by confounding and may reflect a genuine biological causal relationship1. This suggests that SARS-CoV-2 infection could exert lasting effects on central nervous system resilience – potentially through mechanisms such as neuroinflammation or blood-brain barrier disruption – thereby increasing vulnerability to delirium under the physiological stress of surgery7. That said, our MR approach also has limitations. The delirium phenotype in current GWAS is broadly defined and does not distinguish postoperative subtypes or capture timing relative to surgical exposure. Moreover, the genetic instruments for COVID-19 reflect lifetime susceptibility rather than acute infection severity or temporal proximity to surgery. Future studies integrating longitudinal biomarker data – such as neurofilament light chain or inflammatory cytokines – with multicenter perioperative cohorts will be essential to validate the proposed 7-week “neurocognitive safety window” and refine risk prediction across diverse populations7, 8. In conclusion, the clinical insights from Liu et al and the causal inference from our MR analysis are mutually reinforcing: the former delineates a clinically relevant risk pattern, while the latter strengthens its biological credibility1. Together, they underscore the need for clinicians to consider not only pulmonary recovery but also post-COVID brain health when determining optimal timing for elective surgery. Advancing precision prevention of perioperative neurocognitive disorders will require continued integration of observational epidemiology, genetic causal inference, and mechanistic neuroscience.
Qi et al. (2026) studied this question.