Mortality overview

The first page sorts each condition by its pooled risk ratio for death (random effects, DerSimonian–Laird; each trial at its longest timepoint) into lower mortality with albumin, no clear difference and higher mortality (the 95% CI excludes 1), or mortality not tested (no eligible trial, or no trial in that comparison reported mortality). Cirrhosis conditions use albumin against no albumin (SBP, non-SBP infection, hepatic encephalopathy, variceal bleeding) or against other volume expanders (large-volume paracentesis, HRS-AKI); the other pages pool all comparators within each condition. Hover over a condition for its comparison and result; select it to open its forest plot.

Trials and comparisons

Randomised controlled trials with an albumin arm and an albumin-free arm, grouped by clinical indication. Trials in which every arm received albumin (including all HRS-AKI trials) are not plotted. On the cirrhosis pages and the intradialytic hypotension page, comparisons are never pooled across the three comparator categories: albumin against no albumin, against another volume expander, and against a vasoconstrictor. Within a category, each subcomparator is shown alone and all subcomparators are shown combined. On the three Albumin as therapy pages and the albumin plus furosemide page, each condition is analysed separately, and pooled estimates combine trials regardless of comparator (the comparator is shown on every row, and each category or subcomparator can be shown alone). These are the only pages where comparator categories are pooled together. In multi-arm trials, control arms at the same level are combined (events and N summed; means and SDs pooled) and albumin arms are combined before entry.

Sections

Trials outside cirrhosis are grouped by why albumin was given. Albumin as therapy, in three pages: volume resuscitation (albumin compared with other fluids, in sepsis or septic shock, sepsis in cirrhosis (sepsis-induced hypotension), critical illness, traumatic brain injury, ARDS, peri-operative care including cardiac surgery, and burns); treatment of hypoalbuminaemia (albumin given to raise a low serum albumin, compared with no albumin, in sepsis, critical illness, malnutrition on TPN, peri-operative care and cirrhosis, inpatient and outpatient together; no eligible trial was found in nephrotic syndrome, protein-losing enteropathy or cancer); and empiric albumin (albumin given for neither volume nor a low level: stroke, acute pancreatitis and fixed-dose peri-operative albumin). Albumin for volume manipulation: intradialytic hypotension, and albumin given with furosemide, analysed by condition (nephrotic syndrome; cirrhosis, including Gentilini 1999, in which albumin was added to stepped-care diuretics of potassium canrenoate with or without furosemide; and critically ill patients). On the hypoalbuminaemia page, Rubin 1997, Veneman 2004, Yuan 2008 and Lee 2016 are labelled no albumin target: they gave a fixed or weight-based dose without a target serum albumin, in patients selected for hypoalbuminaemia. ANSWER (Caraceni 2018), MACHT (Solà 2018) and Romanelli 2006 are labelled fixed dose, any albumin: long-term fixed doses in outpatients with cirrhosis not selected for a low albumin; MACHT gave albumin with midodrine. Trials on this page whose control arm received other fluids (ALBIOS, ARISS, and Veneman 2004, whose saline and HES arms are combined) are analysed as albumin against no albumin, because both arms received the same background fluids. Trials that used albumin only to prime the bypass circuit are excluded. Pages with no eligible trial are listed with that statement.

Evidence at a glance

Each page opens with a heatmap of pooled estimates, one column per outcome (risk ratios unless the outcome is marked ¶ ratio of means or # mean difference): hard outcomes ({{hard outcomes}}) first, then surrogate outcomes ({{surrogate outcomes}}). Rows are the comparator categories, or on the Albumin as therapy pages an Overall row followed by one row per condition. Each cell uses every trial at its longest timepoint. The colour shows whether the 95% CI excludes no effect and in which direction; a pale tile marks a fragile result (a single trial, or significant by DerSimonian–Laird only, marked *). Some outcomes are not shown on some pages: RRT duration, ventilation duration, ventilator-free days and good functional outcome on the volume resuscitation page, and symptomatic intracranial haemorrhage and infection on the empiric albumin page; their data are kept in the workbook. On the volume resuscitation and hypoalbuminaemia pages serious adverse events are labelled other adverse events. The heatmap footnotes are under the Footnotes dropdown. The Overall row counts each trial once: the SAFE reports on its sepsis (Finfer 2011), traumatic brain injury (Myburgh 2007) and ARDS subsets appear in their own conditions, but the overall estimate uses the whole SAFE trial (Finfer 2004). The same rule applies to the overall estimate in the forest plots. Selecting a cell opens its forest plot.

Timepoints

By default each trial contributes its longest usable timepoint, and the timepoint used is shown on every row (hover it for the wording in the paper). Each timepoint category can also be shown on its own. There is no all-timepoints view, because it would count the same patients twice. When several comparisons or timepoints are selected, each becomes its own plot with its own pooled estimate; they are never pooled together.

Effect measures

Risk ratio for event outcomes. When any cell is zero, 0.5 is added to every cell. When both arms have no events the trial is shown as not estimable and is not pooled.
Ratio of means for urine output, natriuresis and total fluid volume (fluid volumes are reported over different windows and in different units, so a scale-free ratio is used).
Mean difference for length of stay, ventilation duration, ventilator-free days and RRT duration (days).
The forest plot axis covers the point estimates and confidence intervals up to about 0.1–10 (ratios) or ±10 days (mean differences); wider intervals run off the axis and end in an arrowhead.
Medians with IQR or range were converted to mean and SD with the method of Wan et al. (2014). Crossover trials are analysed as parallel-group trials.

Pooling

Random-effects meta-analysis on the log risk ratio, log ratio of means or mean difference, with variance SE². Both confidence intervals shown are random-effects intervals around the same pooled estimate; they differ only in how the interval is calculated. The diamond shows the DerSimonian–Laird 95% CI (normal approximation). The Hartung–Knapp 95% CI (t distribution, k − 1 degrees of freedom) is shown with end caps when there are two or more trials; it is wider when there are few trials. A result is marked fragile when the DerSimonian–Laird CI excludes no effect but the Hartung–Knapp CI does not, or when it rests on a single trial. τ², I² and Cochran's Q are reported for each pooled estimate. On the Albumin as therapy pages, trials are grouped by condition, each with a subtotal, followed by the overall estimate. All pooling is recalculated in your browser when you change the selection.

Direction

Mortality, AKI, RRT, hyponatraemia, paracentesis-induced circulatory dysfunction, ascites recurrence, pulmonary oedema, serious adverse events, postoperative complications, infection, persistent organ failure, rebleeding, symptomatic intracranial haemorrhage, vasopressor use, total fluid volume, ventilation duration, new intradialytic hypotension and length of stay are harmful outcomes, so values below 1 (or below 0) favour albumin. Shock reversal, blood pressure restored after intradialytic hypotension, HE resolution, good functional outcome, urine output, natriuresis and ventilator-free days are beneficial outcomes, so values above 1 (or above 0) favour albumin. Axis labels follow the outcome.

Large-volume paracentesis

Smart 1990, Fassio 1992, Altman 1998 and Zhao 2000 reached large volumes through repeated daily taps of 5 L or less. They are included by default and can be excluded in the forest plot filters.

Albumin-level subgroups

Some trials report results by baseline serum albumin: SAFE (Finfer 2004, subgroups reported in Finfer 2006, ≤25 vs >25 g/L), Park 2019 (≤2.5 vs >2.5 g/dL), Gray 2024 (≤30 vs >30 g/L) and Williams 2025 (<3 vs ≥3 g/dL). These subgroups are shown under their parent trial in the forest plot (open by albumin on the trial row) as hollow circles, and are never pooled, so the totals and pooled estimates are unchanged. Subgroup results were not randomised separately and the thresholds differ, so they carry less weight than whole-trial results. Albumin-level subgroups are not used in the heatmap. No other subgroup analyses are shown. Each forest plot row shows the trial's baseline serum albumin in g/L (the range across arms of the reported mean or median; (entry) = only the entry threshold was reported; NR = not reported); subgroup rows show their threshold. Baseline and achieved serum albumin, and the dosing strategy, are listed in each trial's details.

Intradialytic hypotension (IDH)

Two outcomes are shown. Prevention: albumin given at the start of dialysis (Macedo 2021); the event is a new episode, defined as a fall in systolic blood pressure of 20 mmHg or more during the session. Treatment: albumin given for an established symptomatic episode (Knoll 2004); the event is treatment success, blood pressure restored with up to 750 mL of study fluid (the complement of the treatment failure the trial reports). Macedo 2021 reports per dialysis session rather than per patient; sessions from the same patient are not independent, so its confidence interval is narrower than a patient-level analysis would give. Clark 2021 (SAFER-SLED) was excluded because it is a feasibility (pilot) trial.

Guideline recommendations

Guideline statements on albumin were extracted from six guidelines: AASLD practice guidance (2021), ACG preventive care in cirrhosis (2026), ADQI–ICA on AKI in cirrhosis (2024), ICTMG on intravenous albumin (2024), Baveno VIII (2026) and the Surviving Sepsis Campaign (2026). Strength and certainty were harmonised to GRADE terms (Baveno Oxford levels and ADQI letter grades converted; ungraded consensus statements shown as not graded), and each guideline's native grading system is listed on the Guideline sources page. Each row opens to the statement and any dose. On the Albumin as therapy pages statements are grouped by condition. Supporting text that is not a formal recommendation is listed after the formal statements. Badge colour: green for, red against, grey neutral or context; solid = strong, tinted = conditional, outline = best practice or not graded; circles = GRADE certainty of evidence. The Guideline sources page lists each guideline and the terminology map.

Risk of bias

Risk-of-bias ratings are taken from published systematic reviews and meta-analyses; they were not assessed for this dashboard. The forest plot shows each trial's overall rating: RoB 1 (Cochrane 2011 tool), or RoB 2 when a trial has only RoB 2. The domain ratings of both tools are in each trial's details and on the Trial quality page. A subset report with no rating of its own shows the rating of its parent trial. Where reviews rated the same trial, a Cochrane review's rating takes priority, otherwise the majority rating is used, and ties are shown as unrated; ratings where reviews disagreed are marked ⚑. The Sources column on the Trial quality page lists the reviews behind each trial's ratings.

Data and figures

Every heatmap and forest plot can be downloaded as a figure (PNG), and every forest plot with the trial-level data, the random-effects weights and the source notes, and with the pooled calculations (including condition subtotals), for exactly the selection shown. Data current as of {{data date}}.