Clever physiology, genuine clinical benefit… or just very expensive urine?
Fluid is, on the face of it, reasonably simple.
It goes in. Flows around. It comes out.
Unfortunately, the human body appears to have missed this memo.
Anyone who has spent time looking after patients with heart failure, will be familiar with the slightly bizarre situation where somebody can have oedematous legs, pleural effusions and pulmonary congestion. Yet, they may be hypotensive and have a tongue drier than your dad’s old car washing sponge.
We sometimes describe these patients rather simplistically as being ‘extravascularly wet and intravascularly dry’.
In other words, overloaded in the tissues but with an inadequate effective circulating volume.
Unfortunately, as with most catchy medical phrases, the actual physiology is considerably more complicated. Not every congested patient is literally intravascularly depleted. Venous congestion, cardiac output, renal perfusion, neurohormonal activation and lymphatics are all part of this physiological rave.
But the basic clinical problem remains:
There is fluid in places where we don't want it.
And the kidneys occasionally seem remarkably uninterested in helping.
Enter: Furosemide + Albumin
A medical cocktail that is physiologically appealing enough that I immediately became interested. And according to a recently published randomised trial, it might actually work…
Has somebody finally solved diuretic resistance? Can we cancel renal replacement therapy? Should every ITU immediately fill another fridge with 20% HAS?
Probably not.
But this is a really interesting paper.
Welcome to: Under The Lens
Welcome to the first instalment of Under the Lens.
The part of Four Eyed Medic where I find an interesting paper, stare suspiciously at its methods for far too long, and eventually decide whether I actually believe it.
This isn’t intended to replace a formal critical appraisal. There will be no mandatory PowerPoint or essential pre-reading. And fret not, your attendance will not be required for your portfolio or ARCP.
Instead, I want to answer a few fairly simple questions:
What did they ask?
What did they do?
What did they find?
What’s good?
What made me squint? (pun absolutely intended)
And most importantly:
Would I actually do anything differently because of it?
So, spectacles on. Let’s have a look.
The paper in one minute
The paper is:
Efficacy and safety of human albumin combined with furosemide in acute decompensated heart failure with hepatic dysfunction.
Chalikias and colleagues (2026) conducted the DIORASIS trial. A prospective, single-centre, open-label, randomised pragmatic trial involving 241 patients admitted with acute decompensated heart failure and evidence of hepatic dysfunction. Patients were randomised to either IV furosemide alone, or IV furosemide plus 20% human albumin for the first 72 hours.
Their two co-primary outcomes were:
1) How much the patients felt their symptoms had improved over 72 hours, measured using repeated visual analogue scores
2) The changed in serum creatinine over the same period
The headline result?
Patients receiving furosemide plus albumin reported greater improvement in symptoms and appeared to have more effective decongestion, with less worsening renal function.
Which sounds rather exciting. But we’ll come back to that.
Why should this even work?
First, some physiology.
Albumin is the predominant protein within plasma and an important contributor to plasma oncotic pressure.
If somebody has low serum albumin and a large amount of interstitial oedema, there is therefore an attractively neat hypothesis:
Give albumin.
Increase the oncotic pressure within the vasculature.
Encourage fluid to move from the interstitial space back into the circulation.
Then give a loop diuretic and promptly send that recruited fluid towards the nearest catheter bag.
Problem solved.
There is another potential mechanism too.
Furosemide is highly protein-bound in the circulation and must ultimately reach the tubular lumen to exert its effect. Hypoalbuminaemia has therefore been proposed as one factor that might impair its pharmacokinetics and contribute to diuretic resistance.
So albumin could, theoretically, help both by influencing fluid distribution and by improving the delivery/effectiveness of furosemide.
It's beautiful physiology (or as the Gen-Z readers would say: Chef’s kiss)
And there are few things more dangerous in medicine than physiology that is too beautiful.
Because unfortunately:
Physiological plausibility ≠ clinical benefit.
The human body has an irritating tendency to completely ruin things that worked perfectly well on the whiteboard.
Albumin itself is also not guaranteed to remain neatly inside the vascular compartment, particularly in disease states associated with endothelial dysfunction. The DIORASIS authors themselves point out the theoretical possibility that albumin could extravasate and potentially worsen or prolong extravascular congestion.
So there are good reasons to think this might work.
And some good reasons to worry that it might not.
Which is exactly why we do trials.
What did they actually do?
The investigators screened patients admitted with acute decompensated heart failure requiring decongestion.
To qualify, patients also needed evidence of hepatic dysfunction. This was fairly broad and could include abnormal liver blood tests or ultrasound evidence consistent with hepatic congestion/dysfunction. Patients with liver dysfunction from a non-cardiac cause were excluded.
This is important.
We're not talking about a trial purely involving patients with cirrhosis and severe hypoalbuminaemia.
We're talking primarily about heart failure with evidence of cardiohepatic involvement.
Patients with major haemodynamic instability, severe end-organ hypoperfusion, requirement for inotropes other than digoxin, severe renal dysfunction with an eGFR below 30 mL/min, and some patients with severe pulmonary congestion requiring IV vasodilators were excluded.
So although this paper is extremely interesting from a critical care point of view, it's worth establishing early:
This isn't really an ICU shock trial.
More on that later.
The patients were randomised 1:1. Both groups received a slow continuous intravenous infusion of furosemide.
The intervention group additionally received 20% human albumin, initially 10 g daily, with clinicians able to increase the dose to 15 g on day two and 20 g on day three. Treatment could be adjusted based on clinical response, with clinicians aiming for at least 2.4 litres of urine output per day.
Essentially:
Furosemide group: make them wee
Albumin + Furosemide group: add some expensive protein and then make them wee
Science.
Who were these patients?
There were 241 patients, with a mean age of around 78 years.
The baseline groups were reasonably well matched across most characteristics.
The median left ventricular ejection fraction was around 40%, renal function was moderately impaired on average, and importantly the median serum albumin in both groups was 3.8 g/dL (or 38g/L if you work in the NHS)
Wait.
3.8?
This immediately caught my attention. That’s within the normal range?!
Because the traditional physiological argument for combining albumin and furosemide tends to make most intuitive sense in somebody who is significantly hypoalbuminaemic. These patients, on average, weren't.
That doesn't invalidate the treatment. In fact, it arguably makes the result more interesting.
But it does start raising questions about why any effect occurred.
So, did it work?
On the face of it: Yes.
The primary symptom outcome favoured albumin.
Patients receiving albumin plus furosemide had a greater improvement in their global symptom score over 72 hours, with an AUC of 3,767 compared with 3,457 in the furosemide-only group.P < 0.001.
And it wasn't just one isolated result.
Several secondary measures also pointed towards greater decongestion.
The albumin group lost more weight:
−6.5 kg versus −5.2 kg.
They had greater net fluid loss:
10.7 L versus 9.6 L.
And at 72 hours, 55% were considered free from clinical congestion compared with 30% in the furosemide-only group.
They also reported greater improvement in dyspnoea.
Their median hospital stay was shorter:
4 days versus 5 days.
And worsening renal function occurred in:
25% versus 38%.
At this point you would be forgiven for becoming rather excited.
More fluid off. More weight off. Less congestion. Better symptoms. Fewer episodes of worsening renal function. Home a day earlier.
Frankly, if you stopped reading here, you might think it’s doomsday for the vascath and renal replacement therapy.
But we shall continue reading.
What I liked
There is actually quite a lot to like about this study.
They randomised patients
This sounds obvious, but the previous literature surrounding albumin and loop diuretics has included small and heterogeneous studies. DIORASIS gives us prospective randomised data.
That's useful.
Randomisation was computer-generated, and although the study was open-label to patients and treating clinicians, data collection and outcome assessment involved investigators blinded to treatment allocation.
The groups were reasonably comparable
Baseline characteristics were broadly balanced, which reassures me that randomisation did what it was supposed to do.
They looked beyond urine output
This is a major plus. It is very easy to become overly excited about urine output.
I say this as someone who is a critical care nerd, with an unreasonable obsession about fluid balance. But patients generally do not care whether their urine output improved by another 300 mL. They care whether:
They can breathe
Their legs are less swollen
They feel better
Their kidneys continue working
They get to go home.
DIORASIS included symptoms, clinical congestion, renal outcomes and length of stay. That's much more interesting than simply measuring how enthusiastically the catheter bag filled.
To some people at least.
The furosemide exposure wasn’t dramatically different
The total furosemide administered during the initial 72 hours was broadly similar between groups.
Dismissing the finger-wagging nay sayers who think “well, obviously they got drier, you gave them more furosemide.”
The results point broadly in the same direction
Symptoms. Weight. Fluid balance (yay!). Clinical congestion. Worsening renal function. Length of stay.
When several related outcomes all move in the same general direction, that is more interesting than one solitary positive P-value bobbing around in an otherwise completely negative trial.
And the authors are appropriately cautious in describing this as a hypothesis-generating study.
Which is good. Because now we arrive at the squinting.
What made me squint
1) The main efficacy outcome was subjective... in an open-label study
The primary efficacy outcome was the patient’s own assessment of their symptoms. I actually like patient reported outcomes. Ultimately, making patients feel better is a fundamental part of this medicine business.
The problem isn’t that it’s subjective.
The problem is that patients knew which treatments they were receiving. So did the doctors.
Imagine being told:
“We’re giving you the standard treatment”
Vs
“We’re giving you the standard treatment plus the additional treatment we’re studying”
Could that influence how somebody reports their symptoms? Of course.
It doesn't mean the improvement wasn’t genuine. But it starts ringing alarm bells for an effect which starts with p and rhymes with lacebo…
In fairness, the authors do acknowledge in their limitations that this may have introduced expectation and performance bias to an outcome that depends entirely on what the patient says.
My left eyebrow is now slightly elevated.
2) The treating clinicians knew too
This was designed as a pragmatic trial. I like a pragmatic trial.
They tell us what happens when we do things in something resembling real clinical practice rather than under conditions last encountered during GCSE science.
But pragmatism comes with consequences.
Clinicians knew which arm their patient was in and could adjust the furosemide rate, alter therapy based on clinical response and make other treatment decisions. That creates opportunities for unconscious differences in management.
Even something like length of stay becomes harder to interpret when the clinicians making discharge decisions know who received the intervention.
A one day difference sounds impressive.
But I wouldn't have the discharge lounge order commemorative mugs just yet.
3) A small statistical wrinkle
Now we need to talk about p = 0.045
The trial had two co-primary endpoints: improvement in symptoms and change in serum creatinine.
Because the investigators were testing two primary outcomes, they prespecified a more stringent threshold for statistical significance:
p < 0.025
for each primary endpoint.
For the secondary endpoints, they used the more conventional:
p < 0.05
This isn't necessarily a problem. In fact, using a stricter threshold for multiple primary comparisons is a perfectly reasonable attempt to reduce the chance of finding a positive result purely by chance.
The symptom endpoint?
p < 0.001
Lovely. Well below 0.025. Everyone is happy.
Now the change in creatinine:
Albumin + furosemide: +0.07 mg/dL
Furosemide alone: +0.18 mg/dL
And the P-value?
p = 0.045
Fanfare? Confetti? Usually yes.
However, the authors pre-specified a more stringent p < 0.025 criterion for a primary endpoint. Not a problem, missing a statistical threshold doesn’t make a result useless. But in the results section, the authors describe this difference in creatinine as significant, and it features very strongly in the conclusion.
And here my spectacles required a clean.
I’m reluctant to turn this into a dramatic, Judge Judy style statistical courtroom scene. But I am a stickler for keeping to your word. You told me the line was 0.025. Unfortunately, no amount of statistical optimism makes 0.045 less than 0.025.
The direction of effect is still interesting. The groups did separate and p = 0.045 isn’t suddenly meaningless because somebody drew a line at 0.025. But it does change how strongly I would interpret the finding.
The difference is still potentially clinically meaningful. But it does not appear to fulfil the statistical criterion set by the authors for their co-primary endpoint.
A deflating, yet important distinction.
4) The confidence interval made me squint even harder
Figure 2 adds another small statistical oddity.
The paper reports the between treatment difference in creatinine as around 0.12mg/dL, with a 95% confidence of 0.07 to 0.18mg/dL alongside p = 0.045.
Now I’m not claiming to be the next Prof. Hannah Fry (she is a frankly brilliant mathematician and human if you’re not aware), but those figures are slightly difficult to reconcile.
A 95% confidence interval for the between group difference, that sits comfortably above 0, would normally be expected to correspond to a p value below 0.05. Intuitively rather, more convincingly far below 0.045.
There may be an explanation in how the analyses or intervals were generated. I, of course, don’t have the raw dataset.
As such, as much as I want to, I won’t dramatically stand on a chair and sing out Les Mis style “statistical error!”.
But I would quite like clarification. Which conveniently, is exactly the sort of thing a critical appraisal is supposed to notice.
5) Lots of secondary outcomes means a lot of statistical doors to knock on
The authors tested a fairly large number of secondary efficacy and safety outcomes, generally using a significance threshold of p <0.05. That matters.
Edward Felton (a pretty cool computer scientist) gave us a good quote: “The problem, when you cast your net that wide, is you inevitably catch something you don’t want to catch”. He was talking about computer security research.
In statistics, particularly medical statistics, we often do cast our net wide in the hope that at least one of our findings crosses the magic 0.05 line. In contrast to Ed, Eddie, Edward (I don’t know him well enough to know what he prefers), it is something we do want to catch, though maybe we shouldn’t. It’s that magical threshold we hunt for, and it makes us think “ooh I can publish this”.
I’m not saying the authors planned for this.
For example, the reduction in worsening renal function has a p value around 0.037.
Interesting? Yes. Potentially important? Definitely.
But secondary outcomes with borderline p values deserve a little more caution when many outcomes have been tested. Especially when the corresponding renal co-primary endpoint hasn’t convincingly cleared its own threshold.
6) These patients weren't particularly hypoalbuminaemic
Let’s return to that albumin level.
Median: 3.8 g/dL
In both groups.
Previous evidence has suggested that any diuretic benefit from adding albumin may be greater in patients with more severe hypoalbuminaemia, particularly below about 2.5g/dL, and perhaps with larger albumin doses. A 2021 meta-analysis of 13 studies found greater diuretic and natriuretic effects in patients with lower baseline albumin and higher albumin dosing, while also emphasising substantial heterogeneity in the evidence.
DIORASIS is therefore doing something rather interesting.
The average participant doesn't fit the classic:
“They're very hypoalbuminaemic, therefore let's replace albumin and help the furosemide”
And the authors' analyses suggested that the observed effect wasn't simply explained by the presence or absence of hypoalbuminaemia.
Which makes me wonder whether, if the effect is real, the mechanism has more to do with plasma refill, venous haemodynamics and fluid redistribution than simply correcting an albumin deficiency.
That is physiologically fascinating.
It is also speculation.
This trial wasn't designed to prove the mechanism.
7) "Hepatic dysfunction" covers quite a lot
The title sounds quite specific.
Heart failure with hepatic dysfunction.
But the inclusion definition was relatively broad.
Patients could qualify through abnormal liver blood tests or imaging findings consistent with hepatic dysfunction/congestion. That reflects real life.
Congestive hepatopathy does not politely arrive wearing a badge saying:
Hello, I am clinically significant cardiohepatic dysfunction.
But it does mean the study population is physiologically heterogeneous.
A mildly abnormal GGT in somebody with congestion is not necessarily the same biological problem as substantial hepatic venous congestion with impaired synthetic function.
So identifying which phenotype actually benefits becomes very important.
8) Don't apply this straight to the sickest ITU patient
This is perhaps the most important point for me from a critical care perspective.
The concept is deeply attractive in ITU.
The oedematous patient.
Poor diuretic response.
Kidneys deteriorating.
Fluid everywhere except apparently somewhere useful.
You look at the 20% albumin. You look at the furosemide. You start feeling clever.
But the DIORASIS population excluded patients with significant haemodynamic instability, severe hypoperfusion, severe renal impairment and some forms of severe pulmonary congestion.
So the patient who is overloaded, on vasopressors, oliguric and edging towards renal replacement therapy?
This paper does not tell us that albumin will rescue them from the filter.
Renal replacement therapy manufacturers may stand down from DEFCON 1.
For now.
What we knew before this paper?
Albumin plus furosemide is not a new idea.
A 2014 meta-analysis in hypoalbuminaemic patients found a modest increase in urine output and sodium excretion at eight hours, but those differences were no longer statistically significant at 24 hours. The authors described the overall effect as transient and of modest clinical significance.
The larger 2021 meta-analysis included 13 studies and 422 participants and again found increased urine output and natriuresis with combination therapy, but with substantial heterogeneity between studies. Benefits appeared greater in patients with albumin below 2.5 g/dL and when larger doses of albumin were used.
So before DIORASIS, the evidence could roughly be summarised as “Maybe.”
Which, reassuringly, is one of medicine's most frequently used evidence categories.
Guideline-level evidence has also remained cautious. A 2024 international guideline on intravenous albumin concluded that few routine indications for albumin improve meaningful patient outcomes and generally recommended against its use across several common critical-care scenarios. Though, its evidence review predates DIORASIS.
And that's why this study matters.
It adds a sizeable new randomised trial to a field previously dominated by small, heterogeneous studies.
It moves the evidence forward. It just doesn't finish the argument.
Would I actually use this?
So after all of that:
Has this paper changed my practice?
Not really.
Has it changed how I think?
Absolutely.
I would not currently start routinely giving albumin to every patient with acute decompensated heart failure whose furosemide appears to have gone on strike. There are too many caveats.
The strongest primary finding is subjective in an open-label trial.
The renal co-primary endpoint showed an interesting signal, but did not meet the more stringent significance threshold prespecified for the primary outcomes.
Some of the secondary findings are vulnerable to multiplicity.
This was one centre.
The sickest critically ill patients weren't represented.
And there was no clear signal that this translated into improved 90-day mortality or rehospitalisation outcomes.
But I also don't think the correct response is:
“Pfft. Rubbish study. Ignore it.”
Because there is a remarkably coherent short-term signal.
Patients felt better. They lost more weight. They lost more net fluid. More of them were judged free from congestion. And there was a signal towards less renal deterioration. Something might genuinely be happening here.
The next question is who benefits?
Is it the patient with marked venous congestion?
The person with a particular cardiohepatic phenotype?
The patient with low effective circulating volume?
The hypoalbuminaemic patient?
The diuretic-resistant patient?
Does albumin only work alongside certain contemporary diuretic strategies? And could we identify those patients using better physiological markers rather than simply giving everyone albumin and hoping for the best?
That is the study I'd really like to see.
Large. Multicentre. Preferably placebo controlled. Objective measures of congestion. Modern diuretic protocols. Predefined physiological phenotypes. And patient-centred outcomes.
Which brings us, inevitably, to one of the oldest and most sacred phrases in science:
More research is needed.
Somewhere, a peer reviewer has just shed a tear of joy.
The Four Eyed Scorecard
Clinical interest
5/5 glasses
I love the question.
It tackles a common, frustrating clinical problem with an intervention based on genuinely interesting physiology.
Even if the answer ultimately turns out to be “no”, I want to know.
Methodological strength
3/5 glasses
Prospective. Randomised. Pragmatic.
Reasonable sample for this evidence base.
But single-centre, open-label, and with a subjective primary efficacy outcome vulnerable to expectation bias.
Results
3/5 glasses
There is a persuasive cluster of short-term decongestion findings.
The renal findings are interesting too, although the creatinine co-primary outcome did not cross the trial’s prespecified p < 0.025 threshold, so I would interpret that signal more cautiously.
Applicability
3/5 glasses
Potentially relevant to patients with acute decompensated heart failure and cardiohepatic dysfunction.
Much harder to extrapolate to the haemodynamically unstable, severely renally impaired or critically ill patients in whom it might be most tempting to try it.
Practice changing potential
2/5 glasses
Would I routinely prescribe it tomorrow? No.
Would I now pay considerably more attention when somebody suggests it? Yes.
And perhaps that’s exactly what a good hypothesis generating paper should achieve.
The Four Eyed Verdict
3/5 glasses
This is a genuinely interesting trial.
It asks a clinically relevant question, produces a consistent signal towards improved short-term decongestion and adds an important block of randomised evidence to an area where previous data have been small and messy.
But it is also an excellent example of why reading beyond the abstract matters.
The open-label design matters. The subjective primary outcome matters. The population matters.
And that p = 0.045 sitting next to a prespecified threshold of p < 0.025 really matters.
So:
Would I read it?
Absolutely.
Would I change routine practice based on it?
Not yet.
Has albumin saved us from renal replacement therapy?
Regrettably, no.
Will I now think about this paper every time I've given enough furosemide to make the pharmacist nervous and the patient still refuses to wee?
Almost certainly.
And that, I think, makes it worth putting Under the Lens.
Four Eyed love,
Dr Steff
Enjoyed this? Check out my other posts here!
