Work

Early potassium results for predicting hypokalemia after furosemide in critically ill adults with cardiovascular diagnoses: a multicohort study

Year: 2026
Type: preprint
Study design: Observational Study
Abstract: Background Hypokalemia after furosemide can increase the risk of ventricular arrhythmias in susceptible patients. Earlier prediction studies have not established whether a potassium result available by 12 hours improves prediction of later low potassium or whether that gain holds across hospitals. Methods In a retrospective MIMIC-IV cohort, we compared a baseline logistic model with one adding potassium available by a 12-hour landmark after furosemide. Later hypokalemia was potassium < 3.5 mmol/L from 12 to < 48 hours; non-events required a late test. We assessed the models in later same-hospital records and applied separately fitted, fixed dose-free models to NWICU and INSPIRE using proxy laboratory times. Paired DeLong tests compared assessment-cohort areas under the receiver operating characteristic curve (AUCs); patient-level bootstrap resampling provided 95% intervals. Results Of 2,840 MIMIC-IV patients with cardiovascular discharge diagnoses, 413 had later recorded hypokalemia. Adding early potassium raised the optimism-corrected AUC from 0.686 to 0.745 and lowered the Brier score from 0.1172 to 0.1121. The AUC gain in 953 later same-hospital patients was 0.054 (95% bootstrap interval 0.019–0.089). It was 0.092 in NWICU (n = 369; 65 events; 95% interval 0.018–0.169; P = 0.018), but uncertain in INSPIRE cardiovascular patients (n = 132; 31 events; 0.020, 95% interval − 0.096 to 0.134; P = 0.737). A broader INSPIRE postoperative ICU cohort had a gain of 0.097 (n = 247; 69 events; 95% interval 0.021–0.168; P = 0.012), with underpredicted absolute risk. In MIMIC-IV, 19.2% with an early result lacked enough later testing to classify the outcome. Conclusions Early potassium improved discrimination in selected MIMIC-IV patients. The gain was uncertain in the INSPIRE cardiovascular subgroup, and the original model lacks independent external validation. Prospective testing is needed before using it to guide monitoring. Trial registration Not applicable. This was an observational prediction-model study and no clinical trial was conducted.
Language: English
Subfield: Nephrology
Field: Medicine
Open Access status: green
OpenAlex ID: W7236785730
Cites: 40
Cited by:
Related to: 10