This systematic review and NMA was performed and reported using the Preferred Reporting Items for Systematic Reviews and Network Meta-Analyses (PRISMA) Statement. The protocol was registered with the PROSPERO International Prospective Register of Systematic Reviews (CRD42023476669).

RCTs assessing the efficacy and safety of add-on diuretic therapy currently used in clinical practice—(acetazolamide, sodium-glucose transport-2 inhibitors (SGLT2i), dopamine, mineralocorticoid receptor antagonists (MRAs), thiazides, and tolvaptan)—in the United States alongside loop diuretics in patients hospitalized with acute HF were eligible for inclusion. Loop diuretic as a comparator was classified as no add-on therapy given its current recommended use by clinical practice guidelines.

The primary outcome was hospital length of stay. The secondary outcomes were urine output, weight loss, overall survival, HF rehospitalizations, need for renal replacement therapy (RRT), hypotension, and hypokalemia. The need for RRT was chosen as an outcome over worsening renal function given recent literature showing in the context of good diuretic response, no worsening outcomes in early hospitalization of patients with acute HF.

A literature search of PubMed and EMBASE databases was conducted until March 29, 2024, with no limitations on language. Detailed search strategies for each database are included in the supplemental appendix. A manual search of references along with expert consult was also performed to identify any additional eligible RCTs.

Two reviewers independently assessed the eligibility of RCTs, extracted data on outcomes, and assessed the risk of bias from all included studies. Data was extracted using a standardized data abstraction form. For each study, the study design, location and setting, specific intervention, comparison details, prespecified outcomes, and participant characteristics were collected according to the Cochrane Handbook for Systematic Reviews of Interventions. The risk of bias in individual studies was assessed using the Cochrane Risk of Bias assessment tool for randomized controlled trials (RCTs). The overall certainty of evidence was assessed using the Grading of Recommendations Assessment, Development, and Evaluation (GRADE) method for all direct outcomes. The summary results from the head-to-head comparison of the top 2 regimens as per surface under the cumulative ranking (SUCRA) and any significant indirect comparisons were considered for GRADE assessment. Any discrepancy on data collection was resolved by discussion with a third reviewer.

To evaluate heterogeneity between pooled studies, we calculated an I ². An I ² between 0% and 30% indicates low heterogeneity, 30% to 60% as moderate, and >60% as high.

Dichotomous data was summarized as relative risk (RR), time-to-event as hazard ratio (HR), and continuous data as mean difference (MD) along with 95% confidence intervals (CI). Publication bias was assessed by visual inspection of the funnel plot for symmetry (for outcomes ≥10 RCTs). A random-effects model using the DerSimonian-Laird approach was used to pool studies. Data was pooled using Review Manager software (version 5.4.).

A random-effects network meta-analysis using frequentist methods was applied. Evidence network plots were constructed to show direct comparisons. The thickness of lines between nodes represents the number of direct comparisons. ^, For closed networks, we used node splitting to assess local inconsistency. Heterogeneity was assessed using I ² and Cochrane’s Q statistic. Dichotomous outcomes were summarized as odds ratio (OR), time-to-event outcomes as HR, and continuous outcomes as MD, along with 95% CI. Treatment ranking was estimated using the Surface Under the Cumulative Ranking curve (SUCRA) method. Higher SUCRA values for an individual treatment strategy indicate a superior ranking. Forest plots with the comparisons of the 3 highest-ranked treatments and any other statistically significant comparisons were plotted to display the treatment efficacies of various add-on treatments. The network meta-analysis was performed using the netmeta package in R software (v4.3.1).

The initial search yielded 1,105 references, of which, 29 RCTs published as 33 manuscripts between 2004 and 2024 met the inclusion criteria ( Figure 1 ). ^{,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,} The study characteristics are summarized in Supplemental Table 1 .

Flow diagram depicting the study selection process.

As illustrated in Figure 2 , of the 29 RCTs included, 18 RCTs (62%) were at low risk of bias for the generation of randomization sequence. Thirteen RCTs (45%) were at low risk of bias for allocation concealment. Fourteen RCTs (48%) were at low risk of bias for blinding participants and personnel. Twelve RCTs (41%) were at low risk of bias for blinding of outcome assessment. Twenty-five RCTs (86%) were at low risk of bias for incomplete outcome data. Fifteen RCTs (52%) were at low risk of bias for selective reporting. Eighteen RCTs (62%) were at low risk of bias for other biases, primarily related to stating the use of an intention-to-treat analysis in methods but not following the method in the analysis. For direct meta-analysis, the overall certainty of the evidence ranged from moderate to very low ( Supplemental Table 2 ). For the indirect meta-analysis, the overall certainty of evidence ranged from low to very low ( Table 1 ). None of the visual inspections of the funnel plot indicate a likelihood of overall publication bias for any of the included outcomes.

Risk of bias across included studies.

As shown in Figure 3 , 28 RCTS compared add-on diuretic to no add-on therapy, ^{,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,} and 1 compared SGLT2i to a thiazide. Seven RCTs compared SGLT2i to no add-on therapy, ^,,,,,,, 2 RCTs compared acetazolamide to no add-on therapy, ^,, 2 RCTs compared MRAs to no add-on therapy, ^, 2 RCTs compared thiazides to no add-on therapy, ^, 10 RCTs compared tolvaptan to no add-on therapy, ^,,,,,,,,, and 4 RCTs compared dopamine to no add-on therapy. ^,,, These RCTs enrolled a total of 8,362 patients ( Figure 3 ).

Network diagram for the outcome of (A) Hospital length of stay, (B) Urine output at day 1, (C) Urine output at discharge, (D) Weight loss during hospitalization, (E) Overall survival, (F) HF-rehospitalization, (G) Hypokalemia, (H) Requirement for renal replacement therapy and I) Hypotension. The nodes represent an intervention (e.g. no-add on therapy), and the node size is proportional to the number of patients that received the intervention. The connecting lines across various interventions indicate a direct comparison, and the line’s thickness is proportional to the number of trials with such a direct comparison.

Detailed results including the point estimates, confidence intervals, and heterogeneity for the direct comparisons for all outcomes are summarized in Appendix Table 2 and Appendix Figures 1 through 8. Briefly, add-on therapy was associated with a significant increase in urine output on day 1 and discharge/study completion, weight change during hospitalization and day 1, and a significant reduction in hypokalemia. Furthermore, a sensitivity analysis of low risk of bias RCTs for randomization and allocation concealment ^,,, showed a significantly shorter hospital length of stay with add-on diuretic therapy compared to no add-on therapy (MD −0.52, 95% CI= −1.02,−0.03). The heterogeneity among the pooled RCTs with low risk of bias was low, I ² =0%.