Predicting diuretic resistance without urine output in acute heart failure
E Vice, P Darko, B Otchere, P Berchie, C Chinnatambi, E Hama, B Demoss, A Krishnamoorthy, R Singh, R Loungani, K Bauza, S Damle, E Molina, C MartiAbstract
Introduction
Diuretic resistance is a major barrier to effective decongestion in patients hospitalized with acute decompensated heart failure (ADHF) and is associated with prolonged hospitalization, renal dysfunction, and worse outcomes. Conventional assessments of diuretic response rely on urine output or net fluid balance, yet intake–output documentation is often incomplete outside the ICU. Consequently, diuretic failure may be recognized late, after prolonged ineffective therapy. Early biochemical and physiological changes—such as electrolyte trajectories and vital-sign patterns—may signal impaired renal sodium handling before measurable urine output changes occur. However, these signals are rarely incorporated into prediction models, and many studies exclude patients without complete urine output data, limiting generalizability.
Purpose
To develop and internally validate an early prediction model for diuretic resistance using routinely available data, independent of urine output measurement.
Methods
We conducted a retrospective cohort study of adults hospitalized with ADHF receiving intravenous loop diuretics on general wards. Patients were included regardless of urine output documentation. Predictors available within 48 hours of diuretic initiation included electrolyte trajectories (chloride change, sodium slope), vital-sign trends, weight-based congestion surrogates, and cumulative loop diuretic dose. Urine output and net fluid balance were excluded. Diuretic resistance was defined as failure to achieve estimated −1 L decongestion by 48 hours or need for adjunctive thiazide or acetazolamide therapy. Penalized logistic regression (lasso) and extreme gradient boosting models were developed with cross-validation. Performance was assessed using discrimination and calibration metrics.
Results
The cohort included 5,595 admissions (mean age 71.9 ± 13.6 years; 56.9% male), with furosemide as the index diuretic in 94.2%. Overall, 38.3% met criteria for poor early diuretic response. A model using demographic and early laboratory data demonstrated good discrimination (AUROC 0.745; AUPRC 0.706) and good calibration across risk deciles. Renal function and electrolyte measures were the strongest predictors. Availability of laboratory testing itself added predictive value, suggesting early clinician concern may precede overt diuretic failure.
Conclusion
More than one-third of patients hospitalized with ADHF experienced early diuretic resistance. Using routinely available demographic and laboratory data, diuretic failure was accurately predicted without reliance on urine output. These findings indicate that biological signals of diuretic resistance precede clinical recognition and support earlier, proactive escalation strategies in non-ICU settings. Prospective validation is needed to assess clinical impact.For image description, please refer to the figure legend and surrounding text.For image description, please refer to the figure legend and surrounding text.