Comment on “Association between preoperative hyperglycemia and adverse cardiac events after non-cardiac surgery: a multicenter cohort study”

Article information

Korean J Anesthesiol. 2026;79(4):457-457

Publication date (electronic) : 2026 May 19

doi : https://doi.org/10.4097/kja.26225

Sang Gyu Kwak

Department of Medical Statistics, Daegu Catholic University School of Medicine, Daegu, Korea

Corresponding author: Sang Gyu Kwak, Ph.D. Department of Medical Statistics, Daegu Catholic University School of Medicine, Daegu, Duryugongwon-ro 17-gil 33, Nam-gu, Daegu 42472, Korea Tel: +82-53-650-4724 Fax: +82-53-621-4106 Email: sgkwak@cu.ac.kr

Received 2026 March 3; Revised 2026 April 14; Accepted 2026 May 18.

Dear Editor,

We read with great interest the article by Choi et al. [1], which investigated the association between preoperative acute hyperglycemia and adverse cardiac events following noncardiac surgery using a multicenter Common Data Model database. The large sample size and application of propensity score matching strengthen the study’s contribution to understanding perioperative metabolic risk. Nevertheless, several methodological considerations merit further discussion, as they may affect the validity and interpretation of the reported findings.

First, the exclusion of patients who experienced adverse cardiac events or mortality within the first seven postoperative days may introduce selection bias. Specifically, cohort eligibility is defined using a post-exposure variable, which may result in conditioning on a post-exposure variable. If early postoperative events lie on the causal pathway between preoperative hyperglycemia and subsequent outcomes, their exclusion could distort the estimated exposure–outcome relationship. Moreover, this approach may induce selection mechanisms that alter the observed association, particularly if early complications are causally related to the exposure of interest [2]. Given that early postoperative cardiac events may plausibly represent immediate consequences of perioperative metabolic instability, their exclusion could lead to underestimation of risk. Sensitivity analyses incorporating early events would therefore be valuable in evaluating the robustness of the findings.

Second, the use of a Cox proportional hazards model following 1:2 propensity score matching design requires careful consideration of variance estimation. In matched samples, observations are no longer independent, and appropriate methods such as robust sandwich estimators or stratified Cox models are typically recommended to account for within-matched-set correlation [3]. Failure to account for this dependence may result in underestimated standard errors and artificially narrow CIs, thereby overstating the precision of the estimates. Clarification of whether the analytic approach accounted for the matched design would strengthen confidence in the reported results.

Third, the reported interactions for age and hypertension should be interpreted cautiously. Interaction terms in Cox models are typically assessed on a multiplicative scale and may not directly correspond to clinically meaningful effect modification. From a clinical standpoint, measures on the additive scale or subgroup-specific absolute risks are often more informative for decision-making [4]. Providing subgroup-specific event rates or absolute risk differences could therefore enhance the clinical interpretability of the findings beyond the reported hazard ratios.

In summary, while this study addresses an important clinical question and provides valuable data, consideration of these methodological issues would further strengthen the validity and interpretability of the findings.

­

The authors of the original article were invited to respond to this letter but did not reply. In accordance with our editorial policy, the letter is being published independently to contribute to the ongoing academic discussion.