On the Distribution of Baseline Severity Markers: A Contribution to the Interpretation of Long-Acting Insulin use in Surgical ICU Patients with Type 2 Diabetes

Re: Ammar A, Sorochak N, Paxton R, Lipari M, Giuliano C. Association Between the Use of Long-Acting Insulin and Hypoglycemia in Critically Ill Patients with Diabetes. J Intensive Care Med. 2026;41(5):421–426. doi:10.1177/08850666251371709

To the Editor,

We read with great interest the recent study by Ammar and colleagues ¹ comparing long-acting insulin (LAI) added to sliding-scale insulin (SSI) versus SSI alone in surgical ICU (SICU) patients with type 2 diabetes mellitus (T2DM). The authors are to be commended for addressing a clinically important and underexplored question with a carefully conducted retrospective design. Their findings represent a valuable contribution to the literature, suggesting that basal insulin in this population is not associated with an increased risk of hypoglycemia. We would, however, like to offer a few observations that we believe may enrich the interpretation of the results.

First, a careful inspection of the original study's Table 1 suggests that the majority of baseline severity markers trend numerically higher in the LAI + SSI cohort. While HbA1c (p = .03) and corticosteroid use (p = .03) reach statistical significance — as the authors acknowledge — several additional variables, including baseline glucose (226 vs 190 mg/dL), BMI (28.3 vs 26.7), SOFA score (4 vs 3), AKI stage 1 (23.7% vs 16.7%), AKI stage 2–3 (12.3% vs 8.8%), and vasopressor use (19.3% vs 15.8%), all diverge in the same direction, even where they fall below conventional thresholds for significance. The concurrent, same-direction clustering of these individually modest differences may warrant consideration in the context of confounding by indication. ² It would be reasonable to expect that clinicians might more frequently add LAI to patients perceived as metabolically less stable or harder to control; such a selection mechanism could plausibly account for the residual baseline imbalance observed.

Second, the matching strategy described in the methods rests on duration of inpatient insulin use and pre-admission antihyperglycemic regimen. While clinically intuitive, this approach does not directly incorporate illness severity (eg, SOFA), organ failure, or vasoactive and steroid exposure — variables that are among the most influential determinants of both glycemic outcomes and mortality in the SICU. Propensity score matching or inverse probability weighting incorporating these dimensions could serve as a complementary approach to mitigate residual imbalance in future analyses. ³

Third, in the analysis of secondary outcomes, the authors appropriately apply multivariable logistic regression to mortality and hyperglycemia, which contextualizes the unadjusted signals. A comparable adjustment for ICU length of stay — which remained significantly longer in the LAI + SSI cohort (5 vs 3 days, p = .04) — does not appear to have been presented. Similarly, although the mortality model includes corticosteroid exposure, vasopressor use and acute kidney injury — both of which differed numerically between groups and are well-established ICU mortality predictors ⁴ — were not retained. Additional clarification of how the authors’ own variable-selection criterion (“different between groups and associated with outcomes at p < .1”) was operationalized across these outcomes would assist readers in calibrating their interpretation.

Fourth, the nearly two-fold higher number of hyperglycemic events in the LAI + SSI cohort (1264 vs 663) is interpreted by the authors as reflecting inadequate glycemic control in both arms. An alternative reading is that the LAI + SSI group entered the analysis with a higher glycemic and metabolic burden, and that the mean dose of 0.29 U/kg/day may have been comparatively insufficient for this profile. This interpretation aligns with the trend observed in Table 1 and could motivate future evaluation of dose titration strategies.

Finally, the a priori sample size calculation was powered for the primary hypoglycemia outcome. The adjusted point estimate for mortality (OR 1.6; 95% CI 0.77-3.4) does not exclude a clinically meaningful harm. The breadth of this confidence interval may be more accurately framed as reflecting insufficient power for this secondary outcome rather than evidence of equivalence. ⁵

These observations are offered not to diminish the contribution of the study, but to highlight the uncertainties that may merit consideration when translating the findings into clinical practice. As the authors themselves suggest, prospective and ideally propensity-matched or randomized designs will be essential to more firmly characterize the safety and efficacy profile of LAI + SSI in this population. We thank the authors once again for this important work and welcome their thoughts.

Yours sincerely,