The aim of this study is to evaluate the effectiveness of loop diuretic adaptative algorithm that is based on machine learning, urine output prediction tool, in decongestion of acute heart failure patients.

A total of 90 patients will be enrolled in the study. Of these, 45 will be assigned to the algorithm-based intervention group, while the remaining 45 will serve as the control group. In the control group, all decisions regarding diuretic therapy will be made solely by the attending physician, without the use of the algorithm.

Patients will receive intravenous furosemide, with the initial dose determined by the attending physician. Two hours after administration of the diuretic, a spot urine sample will be collected to measure sodium and creatinine concentrations. Based on these values, the 6-hour urine output will be estimated using the machine learning, urine output prediction tool (http://diuresis.umw.edu.pl). This estimate will guide the diuretic therapy plan for the first 24 hours of hospitalization. On the second day, the procedure will be repeated using the same methodology.

Inclusion Criteria:

• Adult patients over 18 years of age who provide informed consent
• Ability to enroll in the study within the first 24 hours of hospitalization
• Primary reason for hospitalization is acute heart failure with signs of congestion (at least moderate lower extremity edema)
• NT-proBNP > 1500 pg/ml
• Anticipated need for diuretic therapy for at least 48 hours from the time of study enrollment

Exclusion Criteria:

• End-stage kidney disease requiring renal replacement therapy
• Hemodynamic instability requiring inotropic support
• Active infection requiring intravenous antibiotic therapy