Abstract / Summary
Abstract Background : Published DKA risk studies have largely relied on admission biomarkers, organ-failure scores, or multivariable prediction models. These approaches describe baseline severity but do not show whether metabolic acidosis is improving after treatment begins. We examined whether bicarbonate normalization during hours 6–12 after intensive care unit (ICU) admission was associated with discharge alive from the ICU by 72 hours. Methods : This exploratory retrospective landmark study used MIMIC-IV. Adults with DKA were identified using prespecified ICD-9-CM and ICD-10-CM diagnosis codes. The primary cohort required baseline bicarbonate <18 mmol/L, anion gap >12 mmol/L, ICU stay of at least 12 hours, and a bicarbonate measurement during hours 6–12. The exposure was bicarbonate ≥18 mmol/L during hours 6–12. The primary outcome was failure to achieve discharge alive from the ICU by 72 hours, defined as ICU stay ≥72 hours or ICU death. Multivariable logistic regression adjusted for baseline clinical and metabolic variables. Selection associated with repeat measurement was examined with inverse-probability-of-observation weighting. Results: Among 1,243 ICU stays involving adults with DKA, 630 met the active high-anion-gap acidosis phenotype and 569 entered the landmark cohort. Bicarbonate normalized in 261 patients (45.9%). The primary outcome occurred in 52 of 261 patients with normalization (19.9%) and 89 of 308 without normalization (28.9%). In the complete-case model (n=558; 136 events), normalization was associated with lower adjusted odds of the primary outcome (odds ratio [OR], 0.46; 95% confidence interval [CI], 0.28–0.78; P=0.004). The observation-weighted estimate was similar (OR, 0.47; 95% CI, 0.28–0.78). The estimate weakened in the glucose ≥200 mg/dL subgroup and when the last rather than the first bicarbonate value in the initial window was used for adjustment. Conclusions : Bicarbonate normalization during hours 6–12 was associated with a greater likelihood of discharge alive from the ICU by 72 hours. The result was sensitive to the baseline time definition and should be considered hypothesis-generating. Exact laboratory timestamps, detailed treatment data, and external validation are required before clinical use.